Monitoring and Simulation of Hydrology, Suspended Sediment, and Nutrients in Selected Tributary Watersheds of Lake Erie, New York
Links
- Document: Report (49.3 MB pdf) , HTML , XML
- Data Releases:
- USGS data release - Data and rloadest models used to estimate sediment and nutrient loads in selected New York tributaries to eastern Lake Erie
- USGS data release - SWAT Model Archive for Simulation of Hydrology, Suspended Sediment and Nutrients in Selected Tributary Watersheds of Lake Erie, New York
- Download citation as: RIS | Dublin Core
Acknowledgments
The authors would like to thank Allen Young of the U.S. Department of Agriculture Natural Resources Conservation Service for his help in developing the agricultural management and hypothetical modeling scenarios and Rosaleen Nogle of the Buffalo Sewer Authority for providing data for the green infrastructure scenario. We also thank Shannon Dougherty, Karen Stainbrook, and Lauren Townley of the New York State Department of Environmental Conservation for their help with designing the project and compiling the point source data. We thank Raghavan Srinivasan of Texas A&M University for providing a Linux version of the Soil and Water Assessment Tool. Jeff Falgout and Natalya Rapstine of the U.S. Geological Survey (USGS) helpfully provided aided with the use of the USGS Yeti supercomputer. The authors thank Amy M. Russell of the USGS and Raghavan Srinivasan and Sagarika Roth of Texas A&M University for reviewing a draft of this manuscript.
Abstract
The U.S. Geological Survey, in cooperation with Erie County, New York, the New York State Department of Environmental Conservation, and the Great Lakes Restoration Initiative, collected water-quality samples in nine selected New York tributaries to Lake Erie, computed estimates of suspended sediment and nutrient loads using the R scripting package rloadest and used the Soil and Water Assessment Tool (SWAT) to simulate hydrology and suspended sediment and nutrient loads from these tributaries. This project was undertaken to better understand the water quality of New York’s inputs into eastern Lake Erie.
Water-quality samples for suspended sediment, nitrogen, and phosphorus were collected at 19 sampling sites in the Lake Erie Basin in New York. Daily and monthly suspended sediment and nutrient loads were computed with regressions of streamflow and suspended sediment and nutrient concentrations using rloadest.
SWAT models of nine watersheds were created using publicly available data; and the loads were calculated by rloadest. Twenty-six SWAT model scenarios were created to explore the effects that best management practices (BMPs; 21 scenarios), point source discharges (4 scenarios), and green infrastructure (1 scenario) can have on the water quality of the nine tributaries to Lake Erie. BMP scenarios for the watershed models included combinations of agricultural BMPs applied at varying implementation levels across the study watersheds, including cover crops, reduced tillage, nutrient management plans, and filter strips. The BMP scenarios showed small reductions of total nitrogen and total phosphorus. The scenarios have variable suspended sediment load results, with both increases and decreases of sediment modeled. The point source scenarios result in lower total phosphorus loads. The green infrastructure scenario shows only minimal reduction of suspended sediment and nutrient loads from the Buffalo River watershed but shows substantial reductions locally.
Introduction
The U.S. Geological Survey (USGS), in cooperation with Erie County, New York, the New York State Department of Environmental Conservation, and the Great Lakes Restoration Initiative created nine Soil and Water Assessment Tool (SWAT) models of select watersheds in New York within the Eastern Lake Erie Basin.
The section of Eastern Lake Erie Basin in New York has an area of 6,137 square kilometers (km2) and stretches from the border of New York State with Pennsylvania to the confluence of the Niagara River with Lake Ontario. The study area in this report includes nine subwatersheds with a total area of 4,941 km2 (fig. 1) selected for SWAT simulation to represent the Eastern Lake Erie Basin section in New York. These nine subwatersheds capture a broad range of watershed area, topography, land cover, management, soil, and slope types characteristic of the Eastern Lake Erie Basin section in New York. The selected areas are the Big Sister Creek, Buffalo River, Canadaway Creek, Cattaraugus Creek, Chautauqua Creek, Crooked Brook, Eighteenmile Creek, Walnut Creek, and Tonawanda Creek watersheds.
SWAT is a physically based, watershed-scale hydrologic and water-quality model that has been extensively used throughout the United States and the world (Arnold and others, 1998; Douglas-Mankin and others, 2010; Gassman and others, 2007). SWAT uses input land cover, management, soils, elevation, weather, and other data. The SWAT model provides continuous simulation of hydrologic and water-quality processes on a daily time step and permits the assessment of how land-management practices affects water, suspended sediment, and nutrient yields in small or large watersheds with varying soils, land covers, and management conditions over long periods of time. SWAT has widely been used in total maximum daily load applications (Tetra Tech, Inc., 2015), watershed planning (Santhi and others, 2006), and in assessment of best management practices (BMPs; Bosch and others, 2013).
Baseline scenarios were simulated for the 9 watersheds, and an additional 26 scenarios were tested on the 7 calibrated watershed models: 7 low, 7 medium, and 7 high BMP scenarios, 4 point-source discharge limit scenarios, and 1 green infrastructure scenario. SWAT baseline results in this study help identify areas in the study watersheds that contribute large suspended sediment and nutrient loads. The additional scenarios assess how BMP implementation, point-source discharge limits, and addition of green infrastructure may affect suspended sediment and nutrient loads delivered to eastern Lake Erie.
The objective of this study was to better understand the water quality of New York’s inputs into Lake Erie. Specifically, this study provides (1) information regarding the regional hydrologic system and its associated water-quality processes, (2) water resource information that local, State, and Federal entities can use for planning and management purposes, and (3) data that can be used to advance understanding of regional and temporal variations in hydrologic conditions in the study area.
Western Lake Erie has received considerable attention in recent years because of the reemergence of harmful algal blooms that have threatened the drinking water supplies of coastal communities, created a large zone of anoxic water in the lake, and affected shoreline beach and fishery health. Multiple studies have found that agriculture is the leading cause of impairment of waters in Lake Erie (Duncan and others, 2017; Michalak and others, 2013; Scavia and others, 2014; Smith and others, 2015b). Eastern Lake Erie is also stressed by a large population, invasive aquatic species, and large wastewater and agricultural runoff contributions (Buffalo Niagara Riverkeeper, 2014).
A goal of the “U.S. Action Plan for Lake Erie” (U.S. Environmental Protection Agency [EPA], 2018) is for eastern Lake Erie to maintain algae levels below the level constituting a nuisance condition. The majority of historical nuisance benthic algal blooms in eastern Lake Erie were caused by the green algae Cladophora. Cladophora was first documented in the Great Lakes in the 1930s. Since the late 1980s, the extent of Cladophora has steadily been increasing and reaching nuisance levels across the Great Lakes. The most recent recommendations for the Great Lakes Water Quality Agreement Annex 4 (nutrients) subcommittee (Mary Anne Evans, USGS, written commun., 2022) have not set phosphorus loading targets for the Eastern Lake Erie Basin because of the lack of scientific consensus on environmental factors and phosphorus loads and their potential cause to algal blooms in eastern Lake Erie. The Lake Erie Eastern Basin Task Team has instead set out to gather more information on what management efforts are necessary for controlling Cladophora and other algae.
Erie County, on behalf of the Lake Erie Watershed Protection Alliance, in collaboration with the New York State Department of Environmental Conservation (NYSDEC), is developing a nine-element watershed management plan of the Eastern Lake Erie Basin section in New York, which includes the Niagara River Basin, with financial support from the New York Department of State and the Great Lakes Restoration Initiative. A nine-element watershed management plan consistent with guidance from the NYSDEC would identify and quantify sources of pollutants, determine water-quality goals or targets, and describe the BMPs needed to reach said goals or targets.
The SWAT model results in this study may be used by water-resource managers to inform the nine-element watershed plan for the Eastern Lake Erie Basin section in New York. This study may also benefit Federal, State, and county governments and the residents in the study area by providing a quantitative understanding of the sources of nutrients entering streams; assessing the effects of land cover change, BMPs, and point source scenarios; and providing a means to compute suspended sediment and nutrient load estimates for these nine tributaries to the Eastern Lake Erie Basin.
Purpose and Scope
To support the development of a nine-element watershed plan for the New York part of Eastern Lake Erie Basin, the USGS, in cooperation with the NYSDEC, performed water-quality monitoring and developed SWAT models of nine tributary watersheds to Lake Erie. Water-quality monitoring data was regressed against daily streamflow using rloadest, the R package (R Core Team, 2018) for the Load Estimator (LOADEST) regression model, to provide suspended sediment and nutrient loads for SWAT model calibration.
The purpose of this report is to describe water-quality monitoring, loads computation, watershed model development, calibration and validation, and the resulting simulated hydrology and sediment and nutrient loads for nine SWAT watershed models in western New York that drain to Lake Erie (fig. 1). The models were calibrated to streamflow, suspended sediment, phosphorus, and nitrogen loads quantified at USGS streamgages. Modeling scenarios were developed to determine the effect of BMPs on streamflow and water quality. Additional modeling scenarios related to point sources and green infrastructure were explored. Model limitations are discussed.
Map showing the nine study watersheds and simplified hydrology of main stem streams in New York.
Description of Study Area
The study area consists of nine New York watersheds draining into the eastern side of Lake Erie (fig. 1). The selected watersheds are Big Sister Creek, Buffalo River, Canadaway Creek, Cattaraugus Creek, Chautauqua Creek, Crooked Brook, Eighteenmile Creek, Walnut Creek, and Tonawanda Creek (fig. 2). These watersheds are in western New York and together encompass parts of Erie, Niagara, Cattaraugus, Chautauqua, Orleans, Genesee, Wyoming, and Allegany Counties. During the fall of 2017, seven new USGS water-quality streamgages (sites 1, 2, 4, 5, 12–14 in table 1) were installed in the study area. These sites plus 12 existing sites monitor streamflow and water-quality of these watersheds (table 1). Data from 15 streamgages with daily streamflow records (sites 1, 2, 4–6, and 12–21 in table 1) were used to calibrate and validate the SWAT model for streamflow. Data from 13 sites with daily streamflow and water-quality monitoring (sites 1, 2, 4–6, and 12–19 in table 1) were used to create rloadest models of loads that were then used to calibrate and validate the SWAT model for water-quality constituents.
Maps of Soil and Water Assessment Tool basins with locations of U.S. Geological Survey streamgages, concentrated animal feeding operations (CAFOs), National Pollutant Discharge Elimination System (NPDES) point discharges, National Centers for Environmental Information (NCEI) weather stations, and modeled hydrology in the A, Big Sister Creek; B, Buffalo River; C, Canadaway Creek; D, Cattaraugus Creek; E, Chautauqua Creek; F, Crooked Brook; G, Eighteenmile Creek; H, Walnut Creek; and I, Tonawanda Creek watershed models, New York.
Table 1.
U.S. Geological Survey streamflow and water-quality streamgage monitoring sites for the selected tributary watersheds of Lake Erie, New York, examined in this study.[Data are from U.S. Geological Survey (2016a). Baseflow index (BFI) calculated with software from Arnold and Allen (1999) using data in U.S. Geological Survey (2016b). Site numbers correspond to the sites in figure 1. Water-quality data collected include nitrogen, phosphorus, and suspended-sediment concentrations. km2, square kilometer; NY, New York; —, no data; S Br, South Branch; Cr, Creek; Rd, Road]
The studied tributary watersheds are mostly forested, with deciduous forest covering 24.25–69.33 percent of the watershed areas (table 2; fig. 3; National Agricultural Statistics Service [NASS], 2019). There are several concentrated animal feeding operations (CAFOs) in many of the watersheds, the majority being dairies (fig. 2). The outer limits of the City of Buffalo, New York, and its suburbs are near the easternmost end of Lake Erie (fig. 2B). Near the southern lakeshore are small communities and vineyards. There were 43 modeled National Pollutant Discharge Elimination System (NPDES) point source discharges in the watersheds (table 3; fig. 2), 30 of which were from municipal sources.
Maps of Soil and Water Assessment Tool basins with land cover, locations of U.S. Geological Survey streamgages, and modeled hydrology and subbasins of the A, Big Sister Creek; B, Buffalo River; C, Canadaway Creek; D, Cattaraugus Creek; E, Chautauqua Creek; F, Crooked Brook; G, Eighteenmile Creek; H, Walnut Creek; and I, Tonawanda Creek watershed models, New York.
Table 2.
Land cover of selected tributary watersheds of Lake Erie, New York, examined in this study, in 2018.[Land cover data are from the National Agricultural Statistics Service (2019)]
Table 3.
Point sources registered with the National Pollutant Discharge Elimination System of selected tributary watersheds of Lake Erie, New York, examined in this study.[Point-source data and treatment information are from the New York Department of Environmental Conservation (Fisher and Merriman, 2024). Nutrient speciation ratios are from the Chesapeake Bay Program (2010). See model subbasins in figure 3. NPDES, National Pollutant Discharge Elimination System; CO, county; SD, sewer district; STP, sewage treatment plant; St, street; WWTP, wastewater treatment plant; No, number; —, no data]
Physiographically, the Eastern Lake Erie Basin is within the Eastern Lake section of the Central Lowland province of the Interior Plains region, and the Southern New York section of the Appalachian Plateaus province of the Appalachian Highlands region. The area near Lake Erie has lower elevations and shallower slopes than the rolling hills to the east and southeast. The study area watersheds range in size from the very small 13.5 km2 Crooked Brook watershed to the 1,129 km2 Cattaraugus Creek watershed (fig. 1).
There is a size discrepancy between the drainage areas given for the USGS streamgages in table 1 and the SWAT-delineated watershed areas in table 4 because of the following reasons: (1) the different locations used to define a drainage area and (2) different digital elevation models (DEM) used to delineate areas. Firstly, the intersection of the tributary with Lake Erie was used as the SWAT watershed outlet in delineation, whereas the drainage areas in table 1 were delineated at the streamgage location. The streamgages are upstream from the tributary’s confluence with Lake Erie (fig. 2). Secondly, drainage areas corresponding to USGS streamgages in table 1 were delineated using StreamStats (USGS, 2016a), which uses light detection and ranging (lidar) point clouds from the 3D Elevation Program (https://www.usgs.gov/3d-elevation-program) as a DEM. For the SWAT model, the USGS National Elevation Dataset 1/9 arc-second (3.4 meter; https://apps.nationalmap.gov/viewer/) was used as its DEM to delineate watershed areas (table 4). The Big Sister Creek, Chautauqua Creek, and Crooked Brook watersheds delineated for the SWAT models were smaller than the drainage areas determined by StreamStats. The differences in the watersheds’ areas were about 1 km2 or less. This report uses watershed areas delineated using the SWAT model (table 4).
Table 4.
Properties of the study watershed models for selected tributary watersheds of Lake Erie, New York, examined in this study.[Data are from Fisher and Merriman (2024). km2, square kilometer, >, greater than; HRU, hydrologic response unit]
Description of the Study Area Watersheds
Following are physical descriptions of the nine study area watersheds in western New York, including land cover statistics and water quality.
Big Sister Creek Watershed
The 124.3 km2 Big Sister Creek watershed is in Erie County, situated between the Cattaraugus Creek and Eighteenmile Creek watersheds (fig 1). The land cover in this watershed (table 2; fig. 3) is almost half deciduous forest (46.89 percent), with small amounts of agriculture: hay and alfalfa (12.86 percent), corn (7.95 percent), and soybeans (1.95 percent). Vineyards cover a small amount of land on the southwestern side on the watershed (2.51 percent). Other land covers include developed (8.73 percent), woody wetlands (5.65 percent), mixed forest (1.73 percent), and evergreen forest (2.76 percent). There is one CAFO in this watershed (fig. 2). The average slope of the Big Sister Creek watershed is 1.99 percent (table 4). The soils are primarily poorly or very poorly drained (Natural Resources Conservation Service [NRCS], 2019). Tile drainage is used in an estimated 0.87 percent of the total watershed area (table 4; the “Tile Drainage Parameterization” section discusses how tile drainage was estimated).
The NYSDEC (2019) found that the Big Sister Creek watershed is impaired by nutrients, suspended sediment, low dissolved oxygen, and pathogens. Observed nutrient concentrations may be caused by urban and storm runoff, whereas the low dissolved oxygen, pathogens, and suspended sediment may be caused by on-site septic systems.
Buffalo River Watershed
The Buffalo River is formed by the confluence of Buffalo Creek and Cayuga Creek 13.68 kilometers (km) upstream from Lake Erie. Cazenovia Creek joins the Buffalo River 9.27 km upstream from Lake Erie. Cazenovia, Buffalo, and Cayuga Creeks have similar drainage areas, ranging from 30 to 33 percent of the Buffalo River watershed, but the area draining to the streamgages varies by subwatershed (table 1). The largest of these drainage areas is to streamgage 04214500 on Buffalo Creek, that has a drainage area of 368 km2. The drainage area to streamgage 04215500 on Cazenovia Creek is similar in size to the area drained to the Buffalo Creek streamgage, draining 350 km2, whereas the area draining to streamgage 04215000 on Cayuga Creek has a 250 km2 drainage area. Its watershed (1,107.6 km2 area) is primarily in Erie County, and the headwaters of Cayuga and Buffalo Creeks are in western Wyoming County (fig. 1). A small part of the northern part of Cayuga Creek subwatershed is in Genesee County.
The average slope of the Buffalo River watershed is 3.65 percent (table 4). About half of Buffalo Creek and Cazenovia Creek subwatersheds have slopes greater than (>) 5 percent; most slopes (65 percent) in Cayuga Creek subwatershed are less than (<) 5 percent. The Cazenovia Creek and Cayuga Creek subwatersheds soils are primarily poorly or very poorly drained (91 and 80 percent of the total area, respectively) (NRCS, 2019). The Buffalo Creek subwatershed soils are poorly drained (50 percent) and partly well drained and very poorly drained (25 percent each; NRCS, 2019).
The land covers of each these three subwatersheds are similar; however, there is more deciduous forest in the Cazenovia Creek subwatershed (57.3 percent) than Cayuga Creek (42.9 percent) and Buffalo Creek (40.9 percent) subwatersheds (fig. 3B; NASS, 2019). Cayuga and Buffalo Creeks subwatersheds have more row crop agriculture (approximately 12 percent for both watersheds) than Cazenovia Creek subwatershed (2.0 percent). Less tile drainage was estimated in the Cazenovia Creek subwatershed (1 percent of the total area) than in Buffalo Creek subwatershed (13 percent of the total area) and Cayuga Creek subwatershed (14 percent) subwatersheds.
The following impairment data and interpretations are from the NYSDEC (2019). The Buffalo River watershed, from the confluence of Buffalo Creek and Cayuga Creek to Lake Erie, is impaired by polychlorinated biphenyls, low dissolved oxygen, pathogens, and suspended sediment. The causes of the known and suspected impairments are combined sewer overflows, urban runoff, stormwater runoff, industrial inputs, hazardous waste sites, and habitat and hydrologic stream modification. Buffalo Creek, from its source to the confluence of Buffalo Creek and Cayuga Creek, is impaired by suspended sediment and thought to be impaired by nutrients, pathogens, and elevated water temperature. The elevated sediment is from streambank erosion and urban and storm runoff. The elevated nutrients and pathogens are thought to be from agriculture, and the elevated temperature is thought to be from on-site septic systems and road bank erosion. The Cazenovia Creek subwatershed is impaired by pathogens from sewer and septic system discharge and urban and storm runoff. The Cayuga Creek watershed is impaired from pathogens and thought to be impaired from elevated nutrients and suspended sediment, metals, polycyclic aromatic hydrocarbons (PAHs), and low dissolved oxygen. The pathogen impairments are caused by sanitary discharges and the elevated nutrients and suspended sediment, metals, polycyclic aromatic hydrocarbons, and low dissolved oxygen are thought to be caused by on-site septic systems, streambank erosion, urban and storm runoff, and agriculture.
Canadaway Creek Watershed
The Canadaway Creek watershed, which has an area of 101.5 km2, is entirely inside Chautauqua County (fig. 1). Some low-lying areas of the Village of Fredonia near Canadaway Creek have flooded during storms (Lumia and Johnston, 1984). The average slope of the Canadaway Creek watershed is 9.41 percent (table 4). Approximately 65 percent of the watershed is forested, 8.69 percent is developed, 9.00 percent is hay and alfalfa, and 5.49 percent is pasture (table 2). The watershed has several vineyards covering about 6 percent of the watershed, primarily located near Lake Erie. The remaining land cover consists of fruit and vegetable cultivation. It is thought to be impaired by elevated suspended sediment caused by nonpoint sources, logging activities, and natural streambank erosion of highly erodible soils in the watershed (NYSDEC, 2019). Canadaway Creek Wildlife Management Area (https://www.dec.ny.gov/outdoor/82659.html) is in the upstream part of the watershed with steep slopes that may contribute to the natural streambank erosion in the watershed. The average slope of the Canadaway Creek watershed is 9.41 percent (table 4). The majority of soils are poorly drained (58 percent) or very poorly drained (18 percent), whereas a minority of the soils are well or moderately well drained (25 percent; NRCS, 2019).
Cattaraugus Creek Watershed
The Cattaraugus Creek watershed has an area of 1,437.6 km2 (table 1). The headwaters of the Cattaraugus Creek watershed lie in southwestern Wyoming County and northwestern Allegany County, but most of the watershed straddles the boundary of Erie and Cattaraugus Counties (fig. 1). Cattaraugus Creek is part of the county boundary between Cattaraugus and Erie Counties and between Chautauqua and Erie Counties near the watershed outlet to Lake Erie. Part of the Cattaraugus Territory of Seneca Nation of Indians is also in the watershed (fig. 2D). Cattaraugus Creek watershed is the least urbanized of any of the modeled watersheds (fig. 3). Over 50 percent of the watershed is forested, 4.61 percent is developed, and 2.05 percent of the watershed is wetlands (table 2). The remaining land cover of this watershed is agricultural.
Relief of the watershed is 535 meters (m), with an average watershed slope of 9.37 percent (table 4). Upstream areas have steep slopes and wide ridges (NRCS, 2009). Downstream areas have low relief, and the topography ranges from flat to rolling plains. The south side of the watershed has wide, flat valleys with sluggish streams. Over two-thirds of the soils are poorly or very poorly drained (NRCS, 2019). Less than 3 percent of the watershed is estimated to have tile drainage (table 4).
The Cattaraugus Creek watershed is impaired by suspended sediment and nutrients from streambank erosion and agriculture (NYSDEC, 2019). Some of the suspended sediment loading is thought by the NYSDEC (2019) to be natural streambank erosion because of highly erodible soils in the watershed. The Clear Creek subwatershed of the Cattaraugus Creek watershed has no known ecological impairments (NYSDEC, 2019).
Chautauqua Creek Watershed
The Chautauqua Creek watershed is the most southern and western watershed out of the selected study watersheds (fig. 1). Its area is 90.6 km2 (table 1), and it lies entirely in Chautauqua County. Chautauqua Creek watershed has the most forested land cover out of the study watersheds; 69.33 percent of the watershed’s land cover is deciduous forest, 5.90 percent is evergreen forest, and 0.27 percent is mixed forest (table 2; fig. 3E). Less than 5 percent of its area is developed. Hay and alfalfa (8.85 percent) is the largest agricultural land cover in the watershed. There is some vineyard cover (1.22 percent) scattered close to the outlet to Lake Erie. Eighty-four percent of Chautauqua Creek watershed soils are poorly or very poorly drained (NRCS, 2019). This watershed has an average slope of 10.66 percent, the highest out of the studied watersheds (table 4).
The NYSDEC (2019) states that Chautauqua Creek is a water supply for the village of Westfield, N.Y.; agricultural pastureland is suspected of contributing pathogens and causing impairment of Chautauqua Creek.
Crooked Brook Watershed
The Crooked Brook is the smallest watershed (13.4 km2) modeled in this report (fig. 1). This watershed is within Chautauqua County and is adjacent to the Canadaway Creek watershed on its western boundary. The average slope of this watershed is 3.02 percent (table 4). Developed land cover of the city of Dunkirk makes up 42.76 percent of the watershed (table 2). Deciduous forest accounts for 24.25 percent of the land cover. Vineyards are common in the headwaters in the southeastern area of the watershed and near the Crooked Brook outlet, accounting for 14.97 percent of the land cover. Fifty-one percent of the soils in Crooked Brook watershed are well drained (NRCS, 2019). The Crooked Brook watershed is impaired by nutrients caused by sewage waste, municipal and industrial sources, and likely by urban runoff (NYSDEC, 2019). Whereas the other watersheds have streamgages with daily hydrologic data available, only approximately monthly discharge and water-quality measurements were taken at USGS streamgage 0421338405 in the Crooked Brook watershed during this study.
Eighteenmile Creek Watershed
Eighteenmile Creek watershed has an area of 306.8 km2 (table 1) and is entirely in Erie County (fig. 1). Land cover of the Eighteenmile Creek watershed is over 56.38 percent forested (table 2). Developed land covers 8.56 percent of the watershed. The land cover categorized as hay and alfalfa is 13.73 percent. The remaining land cover is of various crops, wetlands, and open water. There are two CAFOs and no point sources in this watershed (fig. 2G). Over 80 percent of the watershed has poorly or very poorly drained soils (NRCS, 2019). Its average slope is 7.48 percent (table 4). The tributary South Branch Eighteenmile Creek drains 94.8 km2 at USGS streamgage 04214231 (table 1; fig. 2G). This streamgage receives streamflow from 31 percent of the watershed.
The Eighteenmile Creek watershed is thought to be impaired by suspended sediment, polychlorinated biphenyls, pathogens, and elevated water temperatures caused by streambank erosion, urban and storm runoff, agriculture, hydrologic modification, and contaminated sediment. However, the South Branch Eighteenmile Creek tributary subwatershed has no known impairments (NYSDEC, 2019).
Walnut Creek Watershed
The Walnut Creek watershed has an area of 130.4 km2 (table 1). Walnut Creek and its tributary Silver Creek join together approximately 0.2 km upstream from the mouth at Lake Erie (fig. 2H). About half of the total watershed area, 60.8 km2, is in the Silver Creek subwatershed. The combined watershed is primarily in Chautauqua County, with a small part in Cattaraugus County (fig. 1).
Deciduous forest is the predominant land cover in this watershed (table 2), with less deciduous forest in the Silver Creek subwatershed (55.3 percent) in comparison to the rest of Walnut Creek watershed (62.4 percent; fig. 3H). The second most common land cover is hay and alfalfa. The Silver Creek subwatershed and the remaining Walnut Creek watershed have a similar area of land in use for vineyards (5.0 percent in Silver Creek subwatershed and 4.2 percent in the rest of Walnut Creek watershed). Soils primarily have a slight slope and are mostly poorly or very poorly drained (NRCS, 2019). Less than 1 percent of the watershed was estimated to have tile drainage (table 4). Average slope of the watershed is 7.84 percent (table 4).
The NYSDEC (2019) found that Silver Creek subwatershed is impaired with low dissolved oxygen and suspended sediment and nutrients caused by local municipal discharges, streambank erosion because of highly erodible soils, logging activities, and other nonpoint sources. The rest of Walnut Creek watershed is impaired by nutrient runoff from agricultural nonpoint sources, streambank erosion, and logging activities (NYSDEC, 2019). This part of the watershed is also thought to be impaired by low dissolved oxygen and suspended sediment which is likely because of highly erodible soils throughout the watershed (NYSDEC, 2019).
Tonawanda Creek Watershed
The Tonawanda Creek watershed is the northernmost of the selected watersheds with an area of 1,630.1 km2 (fig. 1). It is the largest tributary watershed to Lake Erie in New York State (NYS). Its headwaters are in Wyoming County, and Tonawanda Creek flows north through Genesee County and west through Erie County to the Niagara River (fig. 2I). Tonawanda Creek is a part of the Erie Canal (the Erie Canal is locally known as the “New York State Barge Canal”); as such, Tonawanda Creek is regularly dredged from Pendleton, New York, to its mouth at the Niagara River.
Ellicott Creek is the largest tributary to Tonawanda Creek, which begins in the northwest corner of Wyoming County (fig. 2I). Ellicott Creek flows west along the southern boundary of the Tonawanda Creek watershed, and Ellicott Creek joins Tonawanda Creek approximately 0.59 km from its mouth at the Niagara River. Many tributaries of Ellicott Creek have been modified for stormwater conveyance (Buffalo Niagara Riverkeeper, 2014).
The mouth of the Tonawanda Creek is in a developed area (fig. 3I). The eastern part of the watershed is principally agricultural mixed with deciduous forest. Deciduous forest is the dominant land cover of the watershed (24.63 percent), followed by developed (17.35 percent) and woody wetlands (14.92 percent; table 2). The most upstream part of the watershed (primarily in Wyoming County) has steeper slopes. Slopes flatten to the western part of the watershed closer to Lake Erie.
Climate
This study area has a humid continental climate, with heavy climatic influences from Lake Erie to the west and Lake Ontario to the north. Half of the snowfall in winter is caused by early-season lake-effect precipitation. Summers are relatively dry compared to the rest of the northeast United States because the cool water of Lake Erie inhibits storm development (Great Lakes Integrated Sciences and Assessments, undated). Average annual precipitation from 2006 to 2020 was 1,062.23 millimeters (mm) and average annual snowfall was 2,225.04 mm for the city of Buffalo, New York (National Centers for Environmental Information [NCEI], 2023). During the modeling period (2017–19), the average annual temperature ranged from −17.5 to 27.6 °C and the annual precipitation was 1,059 to 1,232 mm, respectively, at the Buffalo Niagara International Airport (Menne and others, 2012).
Geology
The New York part of the Eastern Lake Erie Basin is underlain by bedrock of Silurian and Devonian age and consists primarily of layered shale, limestone, and dolostone (La Sala, 1968). The layers gently dip to the south at about 6–8 meters per kilometer with the oldest bedrock found to the north. In the north, gypsum deposits (calcium sulfate) can be found in the shale, which can also affect that region’s water quality. The shale bedrock comprises mostly black shale, indicating a rich organic origin, and contains elements including iron, manganese, and sulfur that can affect both groundwater and surface-water quality. The limestone and dolostone can locally contribute dissolved calcium and magnesium and a higher pH to groundwater than areas without these rocks.
Soils in the basin are derived from the erosion of the bedrock and deposition of sediments during and following glacial recession. The glacial deposits that overlie the bedrock consist of the following (La Sala, 1968):
-
(1) till, which is a nonsorted and compacted mixture of clay, silt, sand, and gravel deposited directly from the ice sheet to the bedrock surface;
-
(2) lake deposits, which are bedded clay, silt, and sand that settled out in proglacial lakes fed by the melting ice, found in most north-draining valleys in New York; and
-
(3) bedded sand and gravel deposits, which were laid down in the south-draining valleys in New York.
The glacial deposits are generally <15 m thick in the northern study watersheds and in the uplands throughout the study area, whereas thicker glacial deposits are found in the deeply eroded valleys in the southern study watersheds. Postglacial unconsolidated deposits are alluvium which were deposited by streams, and organic wetland deposits formed by accumulation of decayed plant matter in poorly drained areas throughout the basin (La Sala, 1968).
Methods
The following are methods of water-quality data collection, calculation of suspended sediment and nutrient loads, and development of SWAT models, model calibration, and model scenarios to test implementation of BMPs, green infrastructure, and the effect of point sources on water quality. Water-quality samples were collected from 14 tributaries across the nine study watersheds. Subsamples underwent laboratory analysis for constituent concentrations. Daily loads of the constituents were estimated using the R package rloadest and compiled to a monthly total. The estimated monthly loads were compared against simulated loads from the SWAT models. The SWAT model results from calibration and validation periods were statistically analyzed. The SWAT models were then used to test the effects of different BMP combinations and implementation levels on constituent loads.
Water-Quality Monitoring Data Collection
Water-quality samples were collected for concentration analysis of chlorophyll a and pheophytin a, orthophosphate, total phosphorus, total Kjeldahl nitrogen (nitrogen from ammonia and ammonium plus organic nitrogen), nitrate, nitrite, ammonia, suspended solids, suspended sediment, turbidity, and chloride. Field observations were made of water temperature, dissolved oxygen concentration, pH, specific conductivity, and turbidity. There were 361 water-quality samples collected from 14 sites in the 9 study watersheds (sites 1–6 and 12–19 in table 1) approximately monthly between November 2017 and November 2019 (fig. 4). This sample set included 305 regular samples, 35 replicate samples, and 21 blank samples. Sample locations were selected to provide representative coverage of the study watersheds, and to include as much of the drainage area to Lake Erie in New York as possible while remaining above backwater from the lake. Samples were collected at six established streamgages (sites 6, 15, 16, 17, 18, and 19 in table 1), seven newly established streamgages (sites 1, 2, 4, 5, 12, 13, and 14 in table 1), and one small, ungaged tributary close to Lake Erie (site 3 in table 1).
Graphs (A–S) of time series of discharge and water-quality samples collected at study sites on tributaries to Lake Erie, New York.
In 2018, 25 additional samples, including 20 regular samples, 4 replicate samples, and one blank sample were collected from 5 ungaged sites on Cattaraugus Creek and its tributaries (sites 7 to 11 in table 1); these sites were sampled approximately quarterly to investigate variation of water-quality constituents within the Cattaraugus Creek watershed. Discharge was measured at each of these sites when a water-quality sample was collected (fig. 4).
In the first water year (October 2017–September 2018), samples were collected on a regular schedule; in the second year (October 2018–September 2019), high flow events were targeted for sampling. Some scheduled, monthly samples were not collected because of the lapse in Federal appropriation and government shutdown in December 2018. Other scheduled samples were not collected in May and August 2019 because high flows did not occur at a time when crews could sample them. Water-quality samples were collected using the equal-width-increment method and depth integrating isokinetic samplers, as specified in the National Field Manual for Collection of Water Quality Samples (USGS, variously dated), whenever possible. Exceptions to standard sampling protocols were documented on field notes and coded in sample metadata. Low-flow samples were collected by wading (fig. 5), and high flow samples were collected from bridges (fig. 6). Samplers used included the DH-81, DH-95, and D-74AL (Davis and Federal Interagency Sedimentation Project, 2005) except when flow velocities were outside the isokinetic range of the samplers. When velocities were too low, grab samples were collected with open-mouth bottles. When velocities were too high, samples were collected using weighted bottle samplers. Equal-width-increment samples were collected except in the case of rapidly changing conditions during high flow events, when depth integrating or grab samples were collected at the centroids of the left, center, and right channel sections. During sample collection, field measurements of temperature, pH, dissolved oxygen, specific conductivity, and turbidity were made using a multiparameter probe (YSI, Inc. 6-Series Multiparameter Water Quality Sonde) at locations and for durations intended to represent cross-channel conditions. The field observations, water-quality data, methods, and metadata are available in the National Water Information System (USGS, 2016b).
Photograph of low-flow water-quality sample collection with a DH-81 sampler at streamgage 04214500 (site 16 in table 1) on Buffalo Creek, New York, on July 17, 2019. Photograph by Elizabeth Nystrom, U.S. Geological Survey.
Photographs of high-flow water-quality sample collection with a DH-95 sampler at A, streamgage 04214500 (site 16 in table 1) on Buffalo Creek, New York, on April 16, 2018; photograph by Elizabeth Nystrom, U.S. Geological Survey; and B, Streamgage 04215000 (site 17 in table 1) on Cayuga Creek, New York, on April 16, 2018; photograph by Elizabeth Nystrom, U.S. Geological Survey.
Samples were composited in 8-liter plastic churns for splitting into individual bottles for laboratory analysis. After splitting, subsamples for chlorophyll a, pheophytin a, and orthophosphate) analysis were filtered. Subsamples for chlorophyll a and pheophytin a analysis were filtered using a hand-operated vacuum pump and 0.7-micron, glass-fiber filter. Subsamples for orthophosphate analysis were filtered through a 0.45-micron filters using a syringe. Subsamples for analysis of some nutrients, including total Kjeldahl nitrogen (nitrogen from ammonia and ammonium plus organic nitrogen), nitrate, and nitrite, were unfiltered and acidified using sulfuric acid. All subsamples were stored on ice before shipping except samples of suspended sediment (which were unrefrigerated) and chlorophyll a and pheophytin a (which were frozen). Subsamples collected for analyses with short hold times (nutrients) were shipped overnight daily from the field to analyzing laboratories. Subsamples collected for some analyses (suspended sediment, chlorophyll a and pheophytin a) were held and shipped in batches at a later date. Subsamples for chlorophyll a and pheophytin a analysis were shipped on dry ice. Samples were sent to several laboratories for analysis, including ALS (https://www.alsglobal.com/; for nitrite, nitrate plus nitrite, calculated nitrate, total solids, and total dissolved solids), the USGS National Water Quality Laboratory (for total Kjeldahl nitrogen, total phosphorus, chlorophyll a, and pheophytin a), the USGS Soil and Low-Ionic-Strength Water Quality Laboratory (for ammonia, chloride, orthophosphate, and turbidity), and the USGS Kentucky Sediment Laboratory (for suspended sediment).
Development of rloadest Suspended Sediment and Nutrient Load Estimates
The rloadest package (Lorenz and others, 2013; Runkel and De Cicco, 2017), in the R programming language (R Core Team, 2018), was used to evaluate, and when appropriate data were available, to create models to estimate loads from 13 study sites where daily streamflow and water-quality monitoring data were present (sites 1–6 and 12–19 in table 1). The R package rloadest was developed from the Fortran program LOADEST (Runkel and others, 2004). Constituents evaluated for rloadest analysis included total phosphorus, orthophosphate, total nitrogen, nitrate plus nitrite, ammonium, and suspended sediment. For each constituent model, the rloadest program computes regression coefficients by means of the maximum likelihood estimation method (Wolynetz, 1979). For each constituent, three predefined models (Runkel and others, 2004) were tested (table 5), and the models were ranked based on Akaike information criterion scores (Helsel and Hirsch, 2002). Then, diagnostic plots were created to assess the variance (as a function between predicted values and time, season, and discharge) and the normality of each model’s residuals.
Table 5.
The three predefined regression models from rloadest evaluated at each site and when appropriate used to estimate loads of nutrients and suspended sediment.[Equations are from Lorenz and others (2013); Runkel and De Cicco (2017); Runkel and others (2004). lnL, natural log of constituent load; a, coefficient; lnQ, natural log of streamflow minus center of natural log of streamflow; dtime, decimal time minus center of decimal time]
Additionally, the rloadest program computes bias diagnostics that compare estimated constituent loads to observed loads. Load bias percentage is the percentage that the model overestimates (negative number) or underestimates (positive number) the sum of the estimated constituent loads compared to the sum of the observed loads. The partial load ratio is a ratio of the sum of the estimated constituent loads to the sum of the observed loads, which indicates modeled constituent loads were overestimated (>1) or underestimated (<1). The Nash-Sutcliffe efficiency (NSE) is computed by rloadest and provides a measure of model fit that ranges from −∞ (no relation) to 1 (perfect fit). These diagnostics and the graphed residuals were used to select the model that most appropriately estimated loads for each constituent at each site. Models with an inappropriate number of variables compared to the number of samples were not evaluated because of the likelihood of overfitting. In general, 1 model variable (including the intercept) per 10 samples was considered appropriate (Peduzzi and others, 1996).
Once rloadest models were created for each site and constituent, daily mean discharge values could be used to estimate constituent loads. The R package waterData (Ryberg and Vecchia, 2012) was used to screen each site’s discharge record for missing daily mean discharge values or those equal to 0. One of the 13 sites had 1 missing daily mean discharge value; this value was filled with an estimated value using the waterData package fillMiss function. After the discharge records were complete, daily mean discharge was used to estimate loads for each constituent that meets assumptions needed for the rloadest model at each site. Using the adjusted maximum likelihood estimation method, rloadest computed 90-percent prediction intervals (Cohn, 2005). Retransformation bias was automatically corrected by application of a bias correction factor (Bradu and Mundlak, 1970; Cohn, 1988, 2005). All suspended sediment and nutrient load models and estimates are in Bunch (2024).
SWAT Model Development
The SWAT toolbar ArcSWAT 2012 (Texas A&M AgriLife Research, 2022) for the mapping software ArcGIS (Esri, Redlands, Calif.) was used to create the models. SWAT revision 670 was used to model the study watersheds. All spatial data layers were set to the North American Vertical Datum of 1988, with the projection of Universal Transverse Mercator Zone 17N. The DEM used was from the USGS 1/9 arc-second (3.4 m) National Elevation Dataset (USGS, undated). Stream data was taken from the National Hydrography Dataset Plus (USGS, 2019); streams were burned into the DEM as the stream network. The ArcSWAT automatic watershed delineator was used to delineate subbasins within the study watersheds (fig. 3). Subbasin size for each studied watershed was manipulated by changing the location of subbasin outlets so that the area of each subbasin was within an order of magnitude of each other and to approximately match the size of USGS 12-digit hydrologic unit code watersheds (USGS, 2019). Additionally, the USGS streamgages were set as subbasin outlets for model calibration and validation (fig. 2; table 1). The locations of subbasin outlets were manually changed to match outlets of the 12-digit hydrologic unit code watersheds, USGS gages, or to improve spatial precision of important hydrologic features, such as the Erie Canal in the Tonawanda Creek watershed. The watershed outlets for all models were the intersection of the main stream of interest (for example, Cattaraugus Creek of the Cattaraugus Creek watershed) and Lake Erie. The 2018 Cropland Data Layer (CDL; NASS, 2019) was used to provide land-cover data and the Soil Survey Geographic Database (SSURGO; NRCS, 2019) was used to provide soil data. Slope classes were set within ArcSWAT to best represent the topography of each watershed (table 4).
Hydrologic response units (HRUs) are unique topological areas within each subbasin. They are the smallest area unit in SWAT that is independently simulated. HRUs are delineated by the unique combination of subwatershed, land cover type, soil type, and slope class. The combination of these data layers produces thousands of delineated HRUs per study watershed. HRU thresholds were set to a minimum area of 1 hectare (ha), where HRUs smaller than 1 ha were removed and combined with adjacent, larger HRUs to reduce model complexity and processing time. Exceptions to HRU threshold processing were set so that any HRUs with a land cover type classified as septic were not recombined to preserve this land cover type.
SWAT Model Parameterization
SWAT model parameters are changed from default values to represent real-world conditions in a process called parametrization. Model parameters are applied at three different levels: (1) watershed; (2) subbasin; and (3) HRU. Watershed level parameters mostly set the equations or water-quality parameters used throughout the watershed. Groupings of subbasins or HRUs are commonly lumped together to represent similar areas, land covers, management, weather, and so on. Major watersheds are divided into tributary subwatersheds with corresponding subbasins. For example, the Walnut Creek watershed has a tributary of Silver Creek which drains about half of the watershed. Those subbasins that contribute to Silver Creek make the area called “Silver Creek subwatershed” (subbasins 4–7, 9, 10, 13, 16, 17, 21, and 24 in fig. 3H) in this report, and subbasins that contribute to Walnut Creek make the area called “Walnut Creek subwatershed” (subbasins 8, 11, 12, 14, 15, 18–20, 22, 23, and 25–29 in fig. 3H) in this report. One example of HRU groupings is that all forested HRUs throughout a watershed model have the same parameter values for forest land use in this report. SWAT saves its parameters in several different files. Each file type contains multiple parameters; SWAT parameters discussed in this report will take the format of “PARAMETER_NAME.file name.”
Watershed level equations are discussed in the following text; a full description of the following methods used in this study can be found in Neitsch and others (2002). The watershed parameters are stored in the .bsn file. SWAT uses the Soil Conservation Service (SCS) curve number (CN) method (Mockus, 1964) to calculate runoff from HRUs on a daily basis (the U.S. Department of Agriculture [USDA] Soil Conservation Service is now the USDA Natural Resources Conservation Service). In SWAT, HRUs are assigned a CN from 30 to 99; low numbers correspond to low runoff potential, whereas larger numbers correspond to high runoff potential. The runoff CN is a function of slope, land use, soil hydrologic group, soil permeability, and soil moisture. Daily CNs were calculated in the Big Sister Creek and Walnut Creek watersheds as a function of plant evapotranspiration (ICN.bsn set to 1) rather than soil moisture. The other watershed models calculated the CN from soil moisture with adjustments for tile drainage (ICN.bsn set to 2). Potential evapotranspiration was calculated with the Penman-Monteith Method (IPET.bsn set to 1; Neitsch and others 2002) using input weather data, described below. Because all simulations started in January, the models were initialized with snow present in all subbasins (snow_sub.sub set to 150). Channel routing was simulated with the Muskingum method (IRTE.bsn set to 1), as the simulated streamflow performed better than the default variable storage method in most of the watersheds. The new soil phosphorus model (SOL_P_MODEL.bsn set to 1) was used because this algorithm was recommended by White and others (2009) to accurately model phosphorus loads from manure. The instream water-quality model, QUAL2E, is integrated with SWAT to simulate in-stream water-quality processes. QUAL2E simulates nutrient cycles, algae production, oxygen demand and uptake, and atmospheric aeration (Migliaccio and others, 2007). QUAL2E was used in this study by setting the IWQ.bsn parameter to 1.
Collection of Climate Data
Daily precipitation, temperature, and wind speed data from January 1, 1984, to December 31, 2019, were acquired from the National Oceanic and Atmospheric Administration National Centers for Environmental Information weather stations (Menne and others, 2012) for input to the SWAT model (table 6). A subbasin could have different weather stations for precipitation, temperature, wind speed, or relative humidity. The closest weather station for each weather element to each subbasin’s centroid was used (fig. 2). Precipitation and temperature files were processed to replace missing data with values from the nearest weather station. Relative humidity data in New York were obtained from the Iowa Environmental Mesonet (2019). The SWAT weather generator was used to calculate daily values for solar radiation.
Table 6.
Weather stations used in the watershed models for the selected tributary watersheds of Lake Erie, New York, examined in this study.[Global Historical Climate Network (GHCN) data from Menne and others (2012). Relative humidity (RH) data are from Iowa Environmental Mesonet (2019). N, north; S, south; E, east; W, west; NY, New York; PA, Pennsylvania; US, United States; AWOS, Automated Weather Observing System; P, precipitation; Tona, Tonawanda Creek; Buff, Buffalo River; Tmp, temperature; Catt, Cattaraugus Creek; W, wind speed; BgS, Big Sister Creek; E18, Eighteenmile Creek; Cay, Canadaway Creek; Ch, Chautauqua Creek; Cr, Crooked Brook; Walnut, Walnut Creek; NA, not applicable]
Channel and Canal Parameterization
ArcSWAT calculates several parameters for each subbasin, including subbasin area, channel dimensions, and channel slopes from the DEM. SWAT simulates one main channel and one tributary channel per model subbasin saved in the .rte and .sub files, respectively. These parameters, among others discussed in this section, characterize the channels which affect modeled streamflow.
The Manning's roughness coefficient, commonly known as “Manning’s n,” is used in the calculation of channel flow time of concentration—the time it takes for flow from the upstream channels to reach the subbasin outlet. Manning’s n values were selected from Chow (1959) for the study watershed's main stem and tributary channels, represented in SWAT by the parameters of CH_N2.rte for the main stem channels and CH_N1.sub for the tributary channels. Each SWAT subbasin has a Manning’s n value for the main stem channel and the tributary channel. The Manning’s n values for main channels (CH_N2.rte) were set to 0.040 to represent mountain streams with no vegetation and gravels in the channel. For subbasins downstream from USGS streamgage 04213500 on Cattaraugus Creek at Gowanda, N.Y., main stem Manning's n (CH_N2.rte) values were set to 0.025 to represent gravelly stream bottoms. For tributary channels that were unmaintained with dense brush, CH_N1.sub were set to 0.10; this applied to the Buffalo River, Canadaway Creek, Cattaraugus Creek, and Walnut Creek watersheds. The default value of Manning's n, a value of 0.014, was applied to the other watershed tributaries (CH_N1.sub), including Big Sister Creek, Chautauqua Creek, and Eighteenmile Creek watersheds. Flood-control diversions and dredging on Ellicott Creek (Wooster and Matthies, 2008), in the Tonawanda Creek watershed, were represented by setting Manning's n of main channels (CH_N2.rte) to 0.028 (subbasins 63, 64, 67, 70–73, and 76 in fig. 3I).
As a part of the Erie Canal, Tonawanda Creek from Pendleton, N.Y., to Lockport, N.Y., was modified to have a flat hydraulic slope to aid boats in river navigation. From May to October every year, the Lockport lock is opened, which causes the direction of streamflow to be reversed (fig. 2I). Approximately 31.15 cubic meters per second (m3/s) of water flows upstream Tonawanda Creek, exits Tonawanda Creek watershed at Lockport, and continues northeast (Wooster and Matthies, 2008). To simulate the backflow out of the watershed in the Tonawanda Creek model, monthly point sources were added to subbasins 3, 4, 11, 22, 30, 37, and 39 (fig. 3I) that intersect Tonawanda Creek. The average simulated flows during the months of May to October were multiplied by negative 1 to reverse the simulated flow. During months when the Lockport lock would be closed, the point-source flows were set to 0. The main channel slopes (CH_S2.rte) in these Erie Canal subbasins were set to 0.02 meter height per meter width to mimic the flat hydraulic slope of Tonawanda Creek. The main channel Manning's n (CH_N2.rte) of the Erie Canal was set to 0.028 to represent dredged channel bottoms (Chow, 1959).
The Buffalo River from its outlet at Lake Erie to roughly 9.65 km upstream was designated a Federal navigation channel and has been dredged every 2 years by the U.S. Army Corps of Engineers (USACE; USACE, 2010). The depth of this channel has been maintained at 6.7 m beneath the Low Water Datum of Lake Erie (The Low Water Datum of Lake Erie is currently defined as 173.5 m above the 1985 International Great Lakes Datum [USACE, undated b; National Oceanic and Atmosphere Administration, undated]) and the bottom width of the channel was measured as 45.72 m (USACE, 2010). Side slopes have been maintained at a ratio of 1:3 height to width. To simulate the dredged channel bottom, the main stem Manning's n value (CH_N2.rte) for the Buffalo River was set to 0.028 in subbasin 9 (fig. 3B), which encompassed the majority of the Federal navigation channel.
Historically, there were several projects performed to stabilize the banks of Eighteenmile Creek, where rocks were applied to many sections of the banks of Eighteenmile Creek. These projects were done in subbasin 16 (fig. 3G). The main stem Manning's n value (CH_N2.rte) in SWAT was modified to 0.07 for this subbasin.
Groundwater Parameterization
SWAT partitions groundwater into shallow and deep aquifers for each model subbasin. Water in the shallow aquifer contributes water to streamflow (known as baseflow) and water in the deep aquifer is assumed to leave the watershed (Neitsch and others, 2002). To estimate the two SWAT groundwater parameters baseflow alpha factor (ALPHA_BF.gw) and groundwater delay time (GW_DELAY.gw), a baseflow separation algorithm by Arnold and Allen (1999) was used on the daily flow data from streamgage sites 1, 2, 4–6, and 12–21 in table 1. The baseflow alpha factor gives the response of the groundwater to recharge. The groundwater delay time is the number of days for flow to percolate through the soil profile to reach the shallow aquifer—this is a function of hydrologic properties of the shallow aquifer’s geology and the depth to the water table. Other groundwater parameters were set using automatic calibration in the Soil and Water Assessment Tool Calibration and Uncertainty Program (SWAT-CUP); use of SWAT-CUP is described later in the “SWAT Model Calibration and Validation” section.
Management Schedules and Parameterization
All HRUs require a management schedule to define how the land represented by the HRU are used throughout the SWAT simulation. SWAT management schedules can be any length, but they must have at least 1 year. Urban areas, forests, or wetlands typically use the same operations year after year. For example, wetlands are represented in the default management schedule with only two operations per year: the start and end of the growing season. This 1 year of management is then repeated for each year of the SWAT simulation period. Default management schedules for agricultural areas show only the start and end of the growing season for single crop and an autofertilizer which applies a variable quantity of fertilizer depending on crop nutrient needs. Agricultural areas commonly grow different crops in a temporal pattern, referred to as a crop rotation; simulated crop rotations for agricultural areas are discussed in the following section. Additionally, grazing livestock are present within the watersheds. Management schedules for grazing livestock are not a default option within SWAT and their management schedules must be developed. For this report, management schedules will be referred to as “rotations.” For example, the management schedule for forest land cover is “forest rotation.”
The 16 different rotations used in the models are: apple, barren, beef cattle, CAFOs (combining dairy CAFO and poultry CAFO), cash grain, continuous corn, dairy, forest, vineyard, horse, other agriculture, pasture, septic, urban, water, and wetlands. Not every watershed model used every rotation; some land covers in table 2 were not present in all watersheds or not present at great enough quantities to pass the HRU threshold—previously discussed in the “SWAT Model Development” section. The following section described how the rotations were created and applied per HRU so that the management of the study watersheds was simulated as accurately as possible.
For the results reported in this document, some of the similar land covers are lumped together into the same rotation. All forest land covers (deciduous, evergreen, and mixed) are reported together in the forest rotation. The two different wetland types (herbaceous and woody) are reported together in the wetland rotation. The results for the four developed land cover types are reported together in the urban rotation. The “other agriculture” rotation contains the remaining agricultural land covers (oats, potatoes, range, winter wheat; fig. 3); this rotation used all default management schedules and parameters.
Management schedules were created to customize agricultural operations, including fertilizer application quantity and timing, crop type, tillage, and harvest for each crop rotation. To determine the HRUs receiving agricultural rotations, 5 years (from 2014 to 2018) of CDL layers (NASS, 2015, 2016, 2017, 2018, 2019) were combined and the principal crop growing in each HRU was recorded for each of the 5 years. A dairy HRU was defined as a HRU with at least 3 years of alfalfa or pasture and at least 1 year of silage corn or soybeans out of 5 years of CDL data. A cash grain HRU was defined as an HRU with at least 2 years planted with corn or soybeans out of 5 years of CDL data. Continuous corn HRUs were defined as an HRU with at least 3 years of corn out of 5 years of CDL data. Pasture HRUs were defined as an HRU with more than 2 years of pasture or hay and no corn or soybeans grown out of 5 years of CDL data. All other land covers listed in table 2 (excluding corn, soybeans, hay, alfalfa, or pasture) used the management schedule defaults, unless described otherwise in the sections below.
Three management schedules were developed for the following HRUs: (1) dairy HRUs, (2) cash grain HRUs, and (3) continuous corn HRUs. For dairy HRUs, 3 years of corn silage was followed by 5 years of hay in SWAT (table 7). Cash grain HRUs were simulated as 2 years of corn grain followed by 1 year of soybeans (table 8). In continuous corn HRUs, continuous corn rotations were simulated by repeating the schedule of the first year of the cash crop schedule for every year of simulation (table 8). Management schedules for pasture HRUs used SWAT default management.
Table 7.
Simulated conventional and best management practice schedules for dairy rotation used in watershed models for the selected tributary watersheds of Lake Erie, New York, examined in this study.[The data in this table were used for simulating poultry CAFOs after replacing liquid dairy manure of the conventional dairy rotation with poultry litter; poultry litter described in Chiang and others (2010). kg/ha, kilogram per hectare; —, no data]
Incorporated manure application was simulated in the Soil and Water Assessment Tool by setting the parameter FRT_SURF.mgt parameter (fraction of fertilizer applied to the top 10 millimeters of soil) to 0.1.
Subsurface liquid manure injection was simulated in the Soil and Water Assessment Tool by setting the frt_surf parameter (fraction of fertilizer applied to the top 10 millimeters of soil) to 0.01.
Table 8.
Simulated conventional and best management practice schedules for the cash grain and continuous corn rotations used in watershed models for the selected tributary watersheds of Lake Erie, New York, examined in this study.[—, no data; kg/ha, kilogram per hectare; FRT_SURF.mgt, fraction of fertilizer applied to the top 10 millimeters of soil]
For continuous corn rotations, the first year of data in this table is repeated for all following years.
Subsurface manure injections were simulated in the Soil and Water Assessment Tool by setting the FRT_SURF.mgt parameter (fraction of fertilizer applied to the top 10 millimeters of soil) to 0.01.
Side-dressed fertilizer was simulated in the Soil and Water Assessment Tool by setting the FRT_SURF.mgt parameter (fraction of fertilizer applied to the top 10 millimeters of soil) to 0.9.
Broadcast manure applications were simulated in the Soil and Water Assessment Tool by setting the FRT_SURF.mgt parameter (fraction of fertilizer applied to the top 10 millimeters of soil) to 0.95.
The fertilizer (28 percent urea) was represented in the Soil and Water Assessment Tool fertilizer database with the following parameter values: mineral nitrogen (MIN-N) set to 0.280, and mineral phosphorus (MIN-P), organic nitrogen (ORG-N), organic phosphorus (ORG-P), and ratio of ammonia as nitrogen to mineral nitrogen (NH3-N/MIN-N) set to 0.
Management Schedules for Beef Cattle and Horse Rotations
Grazing by beef cattle and horses was simulated in the watersheds. Management schedules for beef cattle are in table 9 and for horses in table 10. The 2017 Census of Agriculture livestock counts by county were downloaded from the USDA National Agricultural Statistics Service (National Agricultural Statistics Service, undated). Livestock were assumed to be evenly distributed across the counties. Horses were simulated on HRUs that were classified with open urban area (URBN) land cover; beef cattle were simulated on HRUs with land cover of rangeland (RNGB) or pasture (PAST). Horses and beef cattle were simulated as grazing from April 15 to October 30, thus the number of grazing days (GRZ_DAYS.mgt) was set to 199 in SWAT. The dry weight amount of manure produced daily by animal (MANURE_KG.mgt) was calculated using values from the standards manual “Manure Production and Characteristics” (American Society of Agricultural Engineers, 2005). For horses, the amount of biomass consumed per animal (BIO_EAT.mgt) was assumed to be 12.5 kilograms (kg) of grass daily, with a 30 percent dry weight, which was assumed to be equal to the amount of biomass trampled (BIO_TRMP.mgt). Following conventional practice, simulated horse manure was accumulated and stored over the winter and was surface applied during the spring. There was another simulated surface application of horse manure in the fall. For beef cattle HRUs, BIO_EAT.mgt and BIO_TRMP.mgt parameters were set to 8 and 3, respectively, using assumptions from Merriman and others (2018b). In the simulations, beef cattle manure was surface applied every 2 weeks during the cold months until grazing began again on April 15.
Table 9.
Simulated conventional and best management practice schedules for the beef cattle rotation in watershed models for the selected tributary watersheds of Lake Erie, New York, examined in this study.[kg/ha, kilogram per hectare; —, no data]
Table 10.
Simulated management schedule for the horse rotation in watershed models for the selected tributary watersheds of Lake Erie, New York, examined in this study.[kg/ha, kilogram per hectare]
Concentrated Animal Feeding Operations
A CAFO is an animal feeding facility that meets animal size thresholds and confines those animals for more than 45 days in any 12-month period. CAFOs are found throughout the study watersheds. Adjustments to HRU management schedules were required to account for high manure production in HRUs that have CAFOs. Customization of management schedules depending on the presence of CAFOs (referred to as “CAFO HRUs”) allowed the SWAT models to accurately simulate the agricultural conditions in the study watersheds. CAFO locations were provided by the NYSDEC and published in Fisher and Merriman (2024) with information on the affected area, quantity of manure applied, and the number and type of animals. For the modeled CAFOs, daily mean manure production and nutrient content were obtained for each animal type from the American Society of Agricultural Engineers (2005). The majority of CAFOs in the study area are dairies.
There is an equine CAFO in the Tonawanda Creek watershed. Based on interviews with USDA personnel, the equine CAFO removes manure from its site, and thus was not considered in this study.
Management Schedules for Dairy Concentrated Animal Feeding Operations
Dairy farms are considered CAFOs if they have 300 or more cows. The CAFOs were overlaid spatially on the HRU framework to determine which HRUs should be simulated as CAFOs. HRUs closest to the CAFO location and designated as dairy or pasture were categorized as CAFOs. HRUs simulated as CAFOs required modifications to the management schedule operations so that the SWAT simulation included the manure rates produced by the CAFOs. Whenever a CAFO location overlapped a dairy rotation HRU, the CAFO manure application schedule for that HRU used the baseline management of dairy rotations shown in table 7, but the application rate of manure was modified to match the CAFO data from the NYSDEC. On pasture HRUs used for dairy cattle grazing, CAFO manure applications were simulated after hay cuttings. The manure application rate was found for each CAFO by dividing the total quantity of manure applied by the CAFO land area.
Management Schedule for Poultry Concentrated Animal Feeding Operations
There are two poultry CAFOs in the study area: one in the Buffalo River watershed and another in the Tonawanda Creek watershed. The poultry CAFO in the Tonawanda Creek watershed was simulated by replacing liquid dairy manure for the conventional rotation in table 7 with poultry litter (droppings mixed with used bedding and spilled feed) in SWAT. Parameters for untreated poultry litter as described in Chiang and others (2010) were added to the SWAT fertilizer database (Neitsch and others, 2002). Based on interviews with USDA personnel, the poultry CAFO in the Buffalo River watershed did not spread any litter on the surrounding agricultural fields, thus poultry litter application was not modeled in that watershed.
Management Schedules for Vineyard Rotations
Management of vineyards (primarily for grape juice production) in the study area was derived from an interview with a local viticulture expert in 2019. Vineyards were simulated by setting the current age of crops (CURYR_MAT.mgt) to 50 years in SWAT, and the simulated management schedule included a fertilizer application in June and biomass harvest in October (table 11). Vineyard rotations were applied on HRUs that have grapes as the land cover.
Table 11.
Simulated management schedule for the vineyard rotation in watershed models for the selected tributary watersheds of Lake Erie, New York, examined in this study.[kg/ha, kilogram per hectare]
| Date | Vineyard rotation schedule |
|---|---|
| May 5 | Start growing season |
| June 15 | 56 kg/ha fertilizer application1 |
| Oct. 10 | Biomass harvest |
This fertilizer (28 percent urea) was added to the Soil and Water Assessment Tool fertilizer database with the following parameter values: mineral nitrogen (MIN-N) set to 0.280, and mineral phosphorus (MIN-P), organic nitrogen (ORG-N), organic phosphorus (ORG-P), and ratio of ammonia as nitrogen to mineral nitrogen (NH3-N/MIN-N) set to 0. The default FRT_SURF.mgt parameter value (fraction of fertilizer applied to the top 10 millimeters of soil) was used, 0.2.
Management Schedules and Parameters for Nonagricultural Rotations
Other land cover types listed in table 2 were simulated in SWAT with the default management schedules, with the exception of apple orchards (apple land cover), forests, and wetlands. The apples orchards, forests, and wetlands were initialized as growing at the beginning of the SWAT simulation period by setting the parameter IGRO.mgt to 1. These were simulated as mature forests and wetlands by setting the current age of trees (CURYR_MAT.mgt) parameter to 30 years for deciduous forests and apple orchards, 10 years for evergreen forests and forested wetlands, 20 years for mixed forest, and 5 years for herbaceous wetlands. Manning's n for overland flow (OV_N.hru), and maximum canopy storage (CANMX.hru) were varied by land cover (table 12). Forest parameters had the largest effect on model results out of all the land cover parameters because of the dominance of forested land cover throughout the watersheds. Recent research indicates that the plant database (plant.dat) parameters are important for model calibration in heavily forested watersheds to properly account for evapotranspiration (Yang and others, 2018, 2019; Yang and Zhang, 2016).
Table 12.
Manning's roughness coefficients and canopy cover parameters by land cover used in watershed models for the selected tributary watersheds of Lake Erie, New York, examined in this study.[SWAT, Soil and Water Assessment Tool]
Septic land cover type was added to the land cover in a process described in the “Septic System Parameterization” section. It is not included in table 2 and figure 3.
The SWAT land-use code refers specifically to the SWAT land-use code used in the SWAT plant growth database (Neitsch and others, 2002).
Tile Drainage Parameterization
Tile drainage was simulated on select agricultural HRUs using the DRAINMOD equations (ITDRN.bsn set to 1) based on the Hooghoudt and Kirkham equations (Moriasi and others, 2013). Tile drainage was assumed when agricultural HRUs (those for dairy, CAFO, cash grain, or continuous corn) had poorly or very poorly drained soils and low slopes (<2 percent). Calculated tile drainage area was variable across the study watersheds, ranging from 0.06 percent in the Canadaway Creek watershed to 10.6 percent in the Tonawanda Creek watershed (table 4). Simulated tile drainage depth (DDrain.mgt) was set to 609.6 mm in SWAT. The depth to impervious layer (DEP_IMP.hru) on tiled HRUs was set to 1,000 mm, beneath the simulated tile drainage depth, to ensure tile drainage is not impeded by the presence of an impervious layer in SWAT (Boles and others, 2015). For HRUs that do not have tile drainage, the parameter DEP_IMP.hru used the default value of 6,000 mm. No pumping from tile drainage was occurring in the basins, thus pump capacity (PC.sdr) parameter was set to 0 in SWAT.
Septic System Parameterization
Septic systems were identified in the study watersheds using the NYS Tax Parcel Centroid Points map layer (New York State Geographic Information Systems [GIS] Clearinghouse, undated), by selecting all points where the sewer description (SEWER_DESC) field was “private.” This layer was overlaid with known water utility service areas to generate a layer of HRUs using septic systems (Fisher and Merriman, 2024). The private septic systems within the water utilities service area were assumed to be hooked up to the municipal wastewater treatment utility and were removed from the septic layer. The remaining septic system points were converted to a raster layer at the same resolution as the land cover layer, 10 m, and then were overlaid onto the land cover layer. This allowed for septic systems to be used as a land cover with the input land cover code of “SEPT.” The septic land covers were excluded from the land cover threshold, meaning the total area of septic land cover was maintained in each subbasin. All HRUs with land cover designation of “SEPT” were modeled as generic conventional septic system (ISEP_TYP.sep set to 1) by activating septic systems on those HRUs (ISEP_OPT.bsn set to 1). Model defaults were used for the rest of the septic parameters. None of the study watersheds have a large presence of septic fields; septic areas are present in 1 percent or less of all study watersheds.
Irrigation and Water Use Parameterization
Site-specific, monthly water withdrawal data from surface water and groundwater sources were available from 2010 to 2018 (NYSDEC, 2020). If more than one water user with the same water source was in the same subbasin, then water use in that subbasin was summed, as SWAT only allows one water user by source type per subbasin. When the water source was from a groundwater, pond, or reservoir source, the monthly average use was compiled by subbasin in the water use files (.wus). Groundwater withdrawals were separated into shallow (WUSHAL.wus) or deep (WUDEEP.wus) aquifers by provided well depth; deep groundwater withdrawals were assumed to be from wells over 30.5 m deep. Point source files were used for withdrawals from streams as they can be exactly specified as a monthly times series (the same format of the observed water use data), rather than using monthly average removals required in the format of the .wus files. Water use withdrawals from streams were entered into the point source files as a negative number to signify a withdrawal.
When the water use of a water user was classified as irrigation, simulated irrigation was applied to HRUs in the management files (.mgt) that corresponded with the location of the source withdrawal. In the management files, the observed monthly water use data was disaggregated into daily irrigation applications, unless irrigation occurred on urban or grassland HRUs. The monthly irrigation volume on urban or grassland HRUs was disaggregated and applied by heat unit scheduling.
Ponds and Wetlands Parameterization
SWAT uses the .pnd files to parameterize ponds and wetlands by subbasin. The U.S. Fish and Wildlife Service National Wetlands Inventory (U.S. Fish and Wildlife Service, 2016) was used to estimate the land area used for the ponds and wetlands. Default parameters were used for all suspended sediment and nutrient related inputs for ponds and wetlands, and other parameters were estimated based on the size of the pond or wetland.
Point Source Parameterization
Monthly point source data from 2013 to 2019 were provided by the NYSDEC (table 3; Fisher and Merriman, 2024). Additional data for some point sources were obtained from the EPA’s Enforcement and Compliance History Online database for the period from 2007 to 2019 (EPA, undated). For point-source data prior to 2007, long-term, monthly averages of the reported data were used. Total nitrogen was assumed to be 21 milligrams per liter (mg/L) where nitrogen data were not reported for a point source; this assumption is based on average nutrient removal efficiencies by various wastewater treatment plant types from Qiu and others (2010). Total phosphorus was assumed to be 3 mg/L and dissolved oxygen was assumed to be 5 mg/L when the respective data was not available for a point source; these assumptions are based on the recommendations of Section 7 of the Chesapeake Bay Phase 5.3 Community Watershed Model (Chesapeake Bay Program, 2010). Total nitrogen was speciated into ammonia, nitrate, nitrite, and organic nitrogen; total phosphorus was speciated into organic and mineral phosphorus (mineral phosphorus was assumed to be orthophosphate for use in SWAT) based on facility type and treatment (Chesapeake Bay Program, 2010). Where quarterly data existed for a given parameter, missing data were filled by interpolating the interquartile range. After data gaps were filled, if more than one discharger was in the same subbasin, then all discharge and effluent loads in that subbasin were summed, as SWAT only allows one point source per model subbasin.
SWAT Model Calibration and Validation
Quantitative statistics including NSE, the coefficient of determination (R2), and percent bias (PBIAS), and visual examination of hydrographs were used to evaluate model performance (Moriasi and others, 2007, 2015). The NSE value ranges from −∞ (no relation) to 1 (perfect fit), where a value of 0.75 to 1 indicates a very good model fit, 0.65 to 0.75 is good, 0.50 to 0.65 is satisfactory, and <0.5 is unsatisfactory. NSE is the most accepted and implemented statistical standard for the evaluation of performance in watershed modeling (Daggupati and others, 2015; Gassman and others, 2007; Mankin and others, 2002) and is well-known for estimating peaks of model responses. R2 is used to evaluate the fit of simulated data to observed data. Higher values of R2 demonstrate a better fit of the simulated data to the observed data. There is only one ranking threshold for R2: an R2 value greater than or equal to (≥) 0.5 is satisfactory. On the other hand, PBIAS is used to assess the average tendency of system response. The optimal value of PBIAS is 0.0 percent, indicating no model bias. Absolute PBIAS values of less than or equal to (≤) 10 percent for flow, ≤15 percent for suspended sediment, and ≤25 percent for nutrients indicate a very good model fit. A good model fit is found when absolute PBIAS values range from 10 to 15 percent for flow, from 15 to 30 percent for suspended sediment, and from 25 to 40 percent for nutrients. Absolute PBIAS values from 15 to 25 percent for flow, 30 to 55 percent for suspended sediment, and 40 to 70 percent for nutrients indicate satisfactory model performance. Absolute PBIAS values ≥25 percent for flow, ≥55 percent for suspended sediment, and ≥70 percent for nutrients indicate unsatisfactory model performance. Positive values of PBIAS indicate that the model underestimates bias, while negative values indicate that the model overestimates bias (Gupta and others, 1999).
Model calibration for hydrology was run on the USGS supercomputer Yeti using SWAT-CUP. SWAT-CUP is an automatic calibration program for SWAT (Eawag, 2013). The Yeti supercomputer uses the Linux operating system (https://www.linux.org/), thus the software must be formatted for the Linux operating system. The SWAT executable file for Linux is available on the SWAT website (https://swat.tamu.edu/software/swat-executables/); the SWAT revision 670 executable file is published in Fisher and Merriman (2024). The SWAT-CUP for Linux program was obtained from 2W2E GmbH (K. Abbaspour, 2W2E GmbH, written commun., 2019). SWAT-CUP was run on Microsoft Windows for suspended sediment and nutrient calibration. Also, sensitivity analysis was performed using SWAT-CUP for Tonawanda Creek watershed and Chautauqua Creek watershed models (app. 1).
Streamflow and water-quality calibration was performed from the beginning of the monitoring periods (shown by site in table 1) until 2018 for the Big Sister Creek, Canadaway Creek, Chautauqua Creek, Eighteenmile Creek, and Walnut Creek watersheds. Longer hydrologic calibration periods were possible in Buffalo River, Cattaraugus Creek, and Tonawanda Creek watersheds because of long-term data. A warmup period of 5 years was used in all models to allow for the models’ initial conditions to stabilize. Water-quality loads used in calibration were computed with rloadest regression with the hydrologic record (see “Development of rloadest Suspended Sediment and Nutrient Load Estimates” section); therefore, the water-quality calibration period coincided with the hydrologic period of record for the Big Sister Creek, Canadaway Creek, Chautauqua Creek, Eighteenmile Creek, and Silver and Walnut Creek watersheds. The Buffalo River, Cattaraugus Creek, and Tonawanda Creek watersheds had longer hydrologic records, so the water-quality calibration period coincided with the water-quality data collection from 2017 through 2018. The model validation period for streamflow and water-quality parameters was from January 1 to December 31, 2019, for all models. SWAT-simulated nitrate as nitrogen loads were calibrated to the rloadest calculated nitrate plus nitrite loads.
Definitions of default values and ranges for model parameters are given in table 13. The calibrated values for these parameters per watershed are in table 14. Groupings of parameter types, such as snow parameters and hydrologic parameters, were calibrated with different methods depending on the nature of the parameter and available data. For the watersheds with more than one USGS streamgage, the model parameters were spatially calibrated to each separate streamgage, starting with parameters at the most upstream streamgage. For example, the Tonawanda Creek watershed model was calibrated for hydrology parameters to four USGS streamgages, three of those on Tonawanda Creek. Parameters for the most upstream streamgage on Tonawanda Creek (streamgage 04216418; Tonawanda Creek at Attica, N.Y.) were calibrated first, then calibration moved downstream to the next streamgage (04217000; Tonawanda Creek at Batavia, N.Y.), and then finally the parameters were calibrated at most downstream streamgage on the Tonawanda Creek at Rapids, N.Y. (streamgage 04218000). Parameters for the fourth streamgage in the Tonawanda Creek watershed on the tributary Ellicott Creek (streamgage 04218518) were calibrated separately. There are three streamgages in the Buffalo River watershed, one each on its main branches: Cazenovia, Buffalo, and Cayuga Creeks. Parameters for the watersheds of each branch were calibrated separately. There are five water-quality monitoring sites in the Cattaraugus Creek watershed, but only one of these has daily streamflow data, streamgage 04213500 on Cattaraugus Creek. Therefore, the Cattaraugus Creek watershed model was calibrated for streamflow at this site. Spatial calibration is indicated in table 14 by specifying different parameter values for different subwatersheds.
Table 13.
Soil and Water Assessment Tool calibration parameters, descriptions, ranges, and default values used for watershed models for the selected tributary watersheds of Lake Erie, New York, examined in this study.[°C, degree Celsius; mm water/(°C×d), millimeter water per degree Celsius multiplied by day; mm, millimeter; HRU, hydrologic response unit; mm/hr, millimeter per hour; mg/L, milligram per liter; m3/Mg, cubic meter per megagram; mg/kg, milligram per kilogram; mg/(m2×d), milligram per square meter multiplied by day]
Table 14.
Soil and Water Assessment Tool calibration values of model parameters for watershed models for the selected tributary watersheds of Lake Erie, New York, examined in this study.[Parameters, default values, and ranges are defined in table 13. Buff, Buffalo River subwatershed; %, percent; Silver, Silver Creek subwatershed; Walnut, Walnut Creek subwatershed; Ell, Ellicott Creek subwatershed; SBR, South Branch Eighteenmile Creek subwatershed ; sub, subbasin; —, no data; Cay, Cayuga Creek subwatershed; Caz, Cazenovia Creek subwatershed; Rapids, portion of Tonawanda Creek subwatershed upstream from streamgage 04218000 and downstream from streamgage 04217000; Bat, portion of Tonawanda Creek subwatershed upstream from streamgage 04217000 and downstream from 04216418; Attica, portion of Tonawanda Creek subwatershed upstream from the Attica streamgage 04216418; Tona, Tonawanda Creek subwatershed; QUAL2E, Enhanced Stream Water Quality Model]
Spatial calibration for the Buffalo River watershed was separated by three main tributaries: Cayuga Creek (Cay), Cazenovia Creek (Caz), and Buffalo Creek (Buff).
Chautauqua Creek watershed was not calibrated for suspended sediment or nutrients. All parameters not listed are defaults.
Spatial calibration for the Eighteenmile Creek watershed was separated into two subwatersheds: Eighteenmile Creek and South Branch Eighteenmile Creek (SBR).
Spatial calibration for the Walnut Creek watershed was separated into two subwatersheds: Silver and Walnut Creeks.
Spatial calibration for the Tonawanda Creek watershed was separated into two or four subwatersheds depending on where data were available. The two subwatersheds were Ellicott Creek (Ell) and Tonawanda Creek (Tona). For the four subwatersheds, this divides the Tonawanda creek subwatershed (Tona) further based on the location of the three streamgages on Tonawanda Creek (Rapids, Batavia [Bat], and Attica) and the existing subwatershed of Ellicott Creek (Ell).
CN2 and SOL_AWC values were increased by the percentage listed. Negative values indicate the values were decreased by the percentage listed.
Snow parameters were first calibrated as recommended by Abbaspour and others (2018) to set an appropriate volume of water in winter months. Snow parameters are typically calibrated at the watershed spatial scale, but some models with multiple streamgages performed differently in the winter months.
Groundwater parameters were calibrated after snow parameters. A baseflow separation (Arnold and Allen, 1999) was applied on streamgages with over 2 years of hydrologic data (table 1) to determine the percentage of flow from groundwater sources. The groundwater parameters baseflow alpha factor, ALPHABF.gw, and groundwater delay time, GW_DELAY.gw, were also estimated by the baseflow separation algorithm. For streamgages with less than 2 years of hydrologic data, ALPHABF.gw and GW_DELAY.gw were set using automatic calibration in SWAT-CUP.
SWAT Model Scenarios
Models created with the previously described calibration parameters in table 13 and 14 are called the baseline scenarios. Further model scenarios tested the following: (1) implementation of selected BMPs, (2) effects from decreasing the phosphorus treatment limits on loads exported from point sources, or (3) the effect of green infrastructure (land use change, permeable pavement, and rain gardens). These three scenario types were developed in collaboration with Erie County and NYSDEC (table 15) and are described in the following sections.
Table 15.
Number and type of simulation scenarios run per study watershed model for the selected tributary watersheds of Lake Erie, New York, examined in this study.[—, no scenario]
Best Management Practice Scenarios
Three BMP implementations scenarios; low, medium, and high; were developed to test the effect of BMPs on watershed streamflow, surface runoff, and water quality. BMPs were implemented on HRUs with the following agricultural rotations: dairy, beef cattle, continuous corn, and cash grain. The simulation specifications of these rotations are described in the section “Management Schedules and Parameterization.” The simulated BMPs are dependent on the type and schedules of simulated agricultural rotations (for example, for the agricultural management schedule of dairy rotation with a BMP applied, see table 7). After simulations were run, data were analyzed to understand the effectiveness the implementation levels and of individual BMPs and BMP combinations on runoff and water quality. No BMP scenarios were run on the Crooked Brook watershed model, which had insufficient streamflow data for calibration, or on the Chautauqua Creek watershed model, which failed calibration and validation (see section “Chautauqua Creek Watershed Model Calibration and Validation”). The low, medium, and high BMP implementation levels were run on agricultural HRUs of the 7 remaining watershed models for a total of 21 BMP model scenarios per agricultural rotation described as follows.
BMPs selected for model scenarios were cover crops (CC), filter strips (FS), reduced tillage (RT), and nutrient management plan (NMP). The selected BMPs were chosen because they were seen as the most likely to be used within the study watersheds. BMPs were modeled with low, medium, and high levels of implementation. The CC, NMP, and RT all occur on the HRU and affect the crop rotation schedules (tables 7, 8, and 9); these will be referred to as infield BMPs, as they occur in the farm field. FS are not infield BMPs; the FS intercepts runoff as it leaves the field and does not disrupt the conventional crop rotation schedules. The medium and high scenarios had multiple BMPs simulated on the same HRU; for the purpose of this report, multiple BMPs will be referred to as BMP combinations. When this occurs, a plus sign (+) is used between the abbreviation of the BMPs to denote multiple BMPs on the same HRU. For example, CC+NMP means that CC and NMP were both applied to the same HRU. The scenarios were as follows:
-
• Low scenario: 10 percent of agricultural areas had a single infield BMP (CC, NMP, or RT) implemented and FS applied to 2.5 percent of agricultural areas;
-
• Medium scenario: the low scenario, plus 10 percent of agricultural areas had two infield BMPs (CC+NMP, CC+RT, or NMP+RT) implemented, and FS applied to 5 percent of agricultural areas; and
-
• High scenario: the medium scenario, plus CC+NMP+RT applied to 10 percent of agricultural areas, and FS applied to 10 percent of agricultural areas.
The low scenario applied either CC, NMT, or RT evenly to a total of 10 percent of the agricultural HRUs in each watershed. The BMPs were applied to randomly selected agricultural HRUs. Each selected infield BMP (CC, NMP, or RT) covers a total of 3.33 percent of the watershed’s agricultural area. FS were simulated on an additional 2.5 percent of the agricultural area and applied to randomly selected agricultural HRUs.
The medium scenario incorporated the low scenario and added combinations of two infield BMPs to an additional 10 percent of the agricultural areas. The three combinations of infield BMPs were the following: CC+RT, CC+NMP, and NMP+RT. The three different BMP combinations each cover 3.33 percent of the watershed’s area. With these 3 combinations and the 3 single BMPs implemented in the low scenario, 20 percent of the agricultural area was simulated with a BMP or BMP combination in the medium scenario. FS were simulated on an additional 2.5 percent of the agricultural areas for a total of 5 percent of the agricultural area.
The high scenario used a combination of three infield BMPs (CC+NMP+RT) on an additional 10 percent of the agricultural areas for a total of 30 percent of the agricultural area with at least one infield BMP applied. FS were simulated on an additional 5 percent of the agricultural area for a total of 10 percent of the agricultural area.
Modifications for BMPs implemented in dairy rotations in comparison to a conventional dairy rotation are shown in table 7. In dairy rotations with CC implemented, the cover crop used was cereal rye. It was planted in October and harvested the following May. When the cover crop was grown, the fall dairy manure application was moved to early October rather than the first of November. The fall tillage was also removed. For dairy rotations with the NMP, the volume of fertilizer applied in the spring of the first and second years was reduced and the manure application method was changed to subsurface injection during in the years with corn silage planted. Subsurface manure injection was modeled by changing the fraction of fertilizer applied to the top 10 millimeters of soil (FRT_SURF.mgt) to 0.01, which applies 99 percent of the manure to the soil subsurface. The side-dress application of fertilizer was removed in the first year and reduced in the second year. Dairy rotations with RT had spring and fall tillage changed to conservation tillage on years with corn silage planted (table 7). Management was not changed years on hay was grown (years 4-8 in table 7) for NMP and RT.
Modifications for BMPs implemented in cash grain and continuous corn rotations in comparison to the conventional rotations are shown in table 8. CC were implemented by planting cereal rye in October and harvesting it in May. For RT implementation, conservation tillage replaced the spring tillage in every year of the rotation. In June of the first and second years of the rotation, the volume of the side-dress fertilizer was reduced, and the fertilizer application method was changed to subsurface injection to simulate NMPs.
No BMPs were modeled on CAFOs as it was assumed that NMPs were already implemented. FS or NMPs were simulated on beef cattle HRUs. NMPs on beef cattle HRUs were simulated by assuming that manure was stored over the winter period instead of applied in biweekly applications. Beef cattle would be inside during cold months (November 1 to April 14), and the facilities were assumed to have manure storage, so manure applications were not simulated during winter (table 9). FS on beef cattle HRUs were modeled using SWAT defaults for all rotation types.
Point Source Scenarios
To find the effect of phosphorus loading to the watersheds from point sources, additional scenarios were simulated by decreasing the limit of total phosphorus in effluent of selected point-source discharges. There were four different point source scenarios: two for the Tonawanda Creek, one for Big Sister Creek, and one for Buffalo River watersheds. Two different total phosphorus limits were used: 0.5 mg/L and 1.0 mg/L. To calculate the total phosphorus loads in the point source scenarios’ inputs, the flow rate was multiplied by the total phosphorus limit and then speciated into mineral and organic phosphorus (see the previous section “Point Source Parameterization”). Occasionally, the calculated total phosphorus load for the scenario would be greater than observed total phosphorus load. In that case, the lower value was used for the scenario. The following are the four point source scenarios (table 3):
-
• The first Tonawanda point-source scenario (Tonawanda scenario 1) changed the total phosphorus limit to 0.5 mg/L on three discharges: NY0025950 (in subbasin 40), NY0026514 (in subbasin 50), and NY0021849 (in subbasin 75; table 3; fig. 3I).
-
• The second Tonawanda point-source scenario (Tonawanda scenario 2) included the settings of scenario 1 and changed the total phosphorus limit to 1 mg/L on three discharges: NY0031003 (in subbasin 45), NY0108430 (in subbasin 59), and NY0020541 (in subbasin 72; table 3; fig. 3I).
-
• The Buffalo point-source scenario changed the total phosphorus limit to 1 mg/L total phosphorus on one discharge: NY0108103 (in subbasin 47; table 3; fig. 3B).
-
• The Big Sister point-source scenario changed the phosphorus limit to 0.5 mg/L total phosphorus on one discharge: NY0022543 (in subbasin 1; table 3; fig. 3A).
Green Infrastructure Scenario
The Buffalo Sewer Authority of the City of Buffalo, N.Y., has heavily invested in green infrastructure, specifically permeable pavement and rain gardens, in the Buffalo urban area. To find the effect of green infrastructure on runoff water quality, a simulated green infrastructure scenario was developed for the Buffalo River watershed. Permeable pavement and rain gardens are in the most downstream subbasin (subbasin 9; fig. 3B). Both these BMPs were modeled in the low-impact development file (.lid) using the fraction of the impervious area that the BMPs covered and default parameters. There were 0.471 ha of permeable pavement and 0.826 ha of rain garden modeled. Additionally, the Buffalo Sewer Authority has demolished vacant buildings and returned those areas to vacant grass lots. The implementation of vacant grass lots were modeled by changing the SWAT land use type (URBLU.mgt), referenced by the urban parameters database (Neitsch and others, 2002), from high density urban to low density residential. This reduced the fraction of impervious area from 0.6 to 0.12 and reduced the runoff CN. The Manning’s n value for overland flow (OV_N.hru) was increased to 0.03 account for the grassed land cover (Chow, 1959). Vacant lots are in subbasins 9, 11, and 20 and cover 2.33, 0.63, and 0.03 km2, respectively.
Calculation of SWAT Model Results
SWAT results are returned on multiple scales and in multiple formats. Suspended sediment and nutrient outputs are given as yields at the HRU and subbasin levels and as loads at the watershed level. A load is the mass of a constituent discharged past a point in a watershed. A yield is the constituent load per unit area of the contributing watershed.
The BMP scenarios were designed to increase the number of BMPs between scenarios by building each successive level of BMP implementation upon the previous scenario. The low scenario applied single BMPs. The medium scenario additionally applied combinations of two BMPs (CC+NMP, CC+RT, NMP+RT). The high scenario additionally applied the combination of three BMPs (CC+NMP+RT). BMPs were applied at the HRU scale; select HRUs had one, two, or three BMPs applied depending on the scenario (low, medium, or high, respectively). An HRU with a BMP in the low scenario has the same BMP applied in the medium and high scenarios. Thus, results at the HRU scale from BMP and BMP combinations from the high scenario can be compiled and analyzed together. Watershed-level results are presented by low, medium, or high scenario as the cumulative effect of all the BMPs combined in each scenario.
After the SWAT scenarios were run, model outputs were analyzed to understand the effectiveness of individual BMPs and BMP combinations on runoff and water quality at the HRU, subbasin, and watershed scales. The average annual suspended sediment and nutrient loads were calculated for the baseline and BMP scenarios at the HRU and watershed scales. Average annual runoff and suspended sediment and nutrient yields were also calculated at the HRU scale. Watershed-scale reductions were calculated by the percent difference in the average annual loads of the test scenario compared to the baseline scenario. Similarly, HRU-scale reductions for BMPs were calculated as the percent difference in yield of the test scenario compared to the baseline scenario.
Results of Data Collection
Results of the water-quality sampling are compiled in boxplots shown in figure 7. Total phosphorus concentrations range from 0.004 to 2.07 mg/L with the maximum value observed at streamgage 04214231 (South Branch Eighteenmile Creek; fig. 7A). Distributions of total phosphorus concentrations show less variability among streamgages compared to orthophosphate concentrations. Streamgage 04218000 (Tonawanda Creek) had the highest median concentration for both total phosphorus (0.094 mg/L; fig. 7A) and orthophosphate (0.015 mg/L; fig. 7B). Total nitrogen ranged from 0.12 to 16.0 mg/L with the maximum value also occurring at streamgage 04214231 (South Branch Eighteenmile Creek; fig. 7C). Median nitrite plus nitrate concentrations were highest at five Cattaraugus Creek streamgages, although only four samples were taken at each of these locations, with a maximum value of 3.06 mg/L from streamgage 04215000 (Cayuga Creek; fig. 7D). Streamgage 04218518 (Ellicott Creek) had both the highest median (0.096 mg/L) and maximum (0.311 mg/L) ammonia concentrations (fig. 7E). The highest median (116.0 mg/L) suspended sediment concentrations were at streamgage 04213500 (Cattaraugus Creek), and the maximum value of 6,690 mg/L suspended sediment was measured at streamgage 04214231 (South Branch Eighteenmile Creek; fig. 7F).
Boxplots of concentrations of A, Total phosphorus; B, Orthophosphate; C, Total nitrogen; D, Nitrate plus nitrite; E, Ammonia; and F, Suspended sediment at the 19 water-quality streamgages (site numbers 1–19, table 1) in New York, from November 2017 to November 2019. Data from U.S. Geological Survey (2016b).
Results of rloadest Models for Suspended Sediment and Nutrient Load Estimates
In total, 46 rloadest models were created for the 13 streamgage sites that had data that met the requirements needed for rloadest analysis. The rloadest results are listed in table 16. Of the 13 sites, rloadest models were created for total phosphorus at 12 sites, total nitrogen at all 13 sites, nitrate plus nitrite at 5 sites, ammonium at 5 sites, and suspended sediment concentration at 11 of the 13 sites. Orthophosphate samples were highly censored and did not produce any viable models at the study sites. Model results that had high bias or nonsignificant model variables were not considered. All model results and associated data are available in Bunch (2024).
Table 16.
Summary statistics of rloadest models used to compute loads at selected U.S. Geological Survey water-quality streamgages.[Data from Bunch (2024). Site numbers correspond to the sites in figure 1. The rloadest regression models are described in table 5. NA, not applicable because there was no rloadest model created; —, no data]
Results of SWAT Model Calibration and Validation
Calculated calibration and validation statistics for the SWAT models are summarized in table 17. SWAT model results were compared to observed streamflow measurements and rloadest calculated constituent loads. The Chautauqua Creek watershed failed calibration; the computed NSE and PBIAS values were unsatisfactory (NSE<0.5; PBAIS≥±25). Crooked Brook watershed was not calibrated because of the lack of long-term daily streamflow measurements. These two models were excluded from table 17 and all model scenarios. The SWAT models generally underestimated monthly average streamflow, monthly suspended sediment loads, and monthly nutrient loads in winter and spring months; these months typically had the largest streamflow and loads, and simulation discrepancies in these months result in lower rankings in calibration statistics. Generally, the growing season (May to September) was well represented. Simulated streamflow was lower in the warmer months than in other months and represented baseflow conditions. Correspondingly, simulated suspended sediment and nutrient loads matched calculated rloadest loads better in warmer months than in colder months. Ammonium was not directly calibrated and is not shown in the following calibration graphs.
Table 17.
Calibration and validation statistics computed on the monthly average streamflow and monthly suspended sediment and nutrient loads for the calibrated Soil and Water Assessment Tool watershed models for selected tributary watersheds of Lake Erie, New York.[Calibration statistics were calculated from simulated results in Fisher and Merriman (2024), observed flow data from U.S. Geological Survey (2016b), and load data from Bunch (2024). Nitrate as nitrogen loads simulated by the Soil and Water Assessment Tool were calibrated to the rloadest calculated nitrate plus nitrite loads (table 16). Crooked Brook watershed model is not included due to lack of daily streamflow data. —, not computed]
Indicates a very good rating according to Moriasi and others (2007). Shaded in green for visibility.
Indicates a good rating according to Moriasi and others (2007). Shaded in yellow for visibility.
Indicates a satisfactory rating according to Moriasi and others (2007). Shaded in gray for visibility.
Indicates an unsatisfactory rating according to Moriasi and others (2007).
Big Sister Creek Watershed Model Calibration and Validation
The Big Sister Creek watershed SWAT model was calibrated for monthly average streamflow and monthly suspended sediment and monthly total nutrient loads at USGS streamgage 04214060, 3.91 km from the watershed outlet (table 17; fig. 8). NSE statistics were rated good for monthly streamflow and monthly suspended sediment, total phosphorus, and total nitrogen loads (table 17). PBIAS ratings for monthly suspended sediment and monthly total nutrient loads are very good, whereas the PBIAS for streamflow were rated satisfactory (table 17). All R2 values are ≥0.68 (table 17), except for ammonium which has an R2 value of 0.48. The simulated flow was underestimated during peak flows (fig. 8). During the largest peak in streamflow during the simulation period (November 2017), the model underestimated suspended sediment and nutrient loads. The nitrate plus nitrite loads estimated by rloadest were not statistically significant and were not calibrated.
Graphs comparing Soil and Water Assessment Tool simulated monthly streamflow and loads to observed monthly streamflow or rloadest calculated loads from January 2017 to December 2019 at U.S. Geological Survey streamgage 04214060 on Big Sister Creek, in the Big Sister Creek watershed, New York, of A, Average streamflow; B, Suspended sediment load; C, Total phosphorus load; and D, Total nitrogen load.
As shown in table 17, model performance in the validation period was not rated as highly as in the calibration period. Only the NSE value for monthly average streamflow was rated good. The NSE statistics for the other estimated constituents (monthly suspended sediment and monthly total phosphorus and nitrogen loads) were rated unsatisfactory. PBIAS, however, was rated good for monthly suspended sediment, monthly total nitrogen, and monthly total phosphorus load values. The model overestimated the monthly suspended sediment and monthly total nutrient loads in the month of January 2019 (fig. 8). An ice jam was reported on Big Sister Creek on February 5, 2019 (USACE, undated a); the breakup of ice jams can mobilize significant amounts of suspended sediment and nutrients.
Buffalo River Watershed Model Calibration and Validation
The Buffalo River watershed SWAT model was calibrated to the streamgages on its three main tributaries. For the model calibration at the Cazenovia Creek at Ebenezer, N.Y. (streamgage 04215500), and Cayuga Creek near Lancaster, N.Y. (streamgage 04215000), streamgages, the NSE and PBIAS statistics were rated good and very good for streamflow (table 17). The streamflow had satisfactory NSE and good PBIAS ratings at streamgage 04214500, the Buffalo Creek streamgage. The negative PBIAS value at streamgage 04215000 indicates that streamflow was overestimated, whereas the positive PBIAS values at the other two tributary streamgages in this watershed show that simulated streamflow was underestimated compared to observed values. Winter flows were generally underestimated (figs. 9, 10, 11), with the exception of flows during January 2018 at 04214500 and 04215500 streamgages and both January and February 2018 at streamgage 04215000. Peaks in streamflow and loads in November 2017 at all three streamgages were underestimated by the models.
Graphs comparing Soil and Water Assessment Tool simulated monthly streamflow and loads to observed monthly streamflow or rloadest calculated loads from January 2017 to December 2019 at U.S. Geological Survey streamgage 04215500 on Cazenovia Creek, in the Buffalo River watershed, New York, of A, Average streamflow; B, Suspended sediment load; C, Total phosphorus load; and D, Total nitrogen load.
Graphs comparing Soil and Water Assessment Tool simulated monthly streamflow and loads to observed monthly streamflow or rloadest calculated loads from January 2017 to December 2019 at U.S. Geological Survey streamgage 04214500 on Buffalo Creek, in the Buffalo River watershed, New York, of A, Average streamflow; B, Suspended sediment load; C, Total phosphorus load; and D, Total nitrogen load.
Graphs comparing Soil and Water Assessment Tool simulated monthly streamflow and loads to observed monthly streamflow or rloadest calculated loads from January 2017 to December 2019 at U.S. Geological Survey streamgage 04215000 on Cayuga Creek, in the Buffalo River watershed, New York, of A, Average streamflow; B, Suspended sediment load; C, Total phosphorus load; and D, Total nitrogen load.
Monthly suspended sediment load NSE calibration values at the streamgages 04214500, 04215500, and 04215000 were rated unsatisfactory, good, and satisfactory (table 17), respectively. At streamgages 04214500 and 04215500, monthly suspended sediment load calibration PBIAS values were rated good. Negative PBIAS values indicate that simulated monthly suspended sediment loads were overestimated at these two streamgages. The streamgage 04215000 PBIAS rating was rated very good (1.70 percent) for suspended sediment and was only slightly underestimated. R2 for monthly suspended sediment load at 04214500 was rated unsatisfactory.
At two of the three tributary streamgages, NSE statistics of simulated monthly total phosphorus loads were rated good in the calibration period (table 17). PBIAS values at all three streamgages were rated good or very good for monthly total phosphorus loads. The R2 values for simulated monthly total phosphorus loads were satisfactory. The NSE and PBIAS statistics for the simulated total phosphorus loads at streamgage 04215000 were rated unsatisfactory and very good, respectively. The NSE ratings for monthly total phosphorus loads at streamgages 04215500 and 04214500 were good. PBIAS values were rated very good for simulated monthly total phosphorus loads at streamgages 04214500 and 04215000. The calculated PBIAS statistic for simulated total phosphorus loads at streamgage 04214500 was negative, indicating that the model overestimated total phosphorus loads at this location, whereas the PBIAS value was positive for simulated total phosphorus at streamgage 04215000, indicating that the model underestimated total phosphorus loads at this streamgage.
NSE statistics for monthly total nitrogen load calibration were rated at least satisfactory and PBIAS statistics were rated good or very good at all three streamgages in the Buffalo River watershed (table 17). All R2 values were rated satisfactory for total nitrogen. The NSE, PBIAS, and R2 statistics for ammonium calibration data at streamgage 04214500 were all rated unsatisfactory.
In the validation period, the NSE statistics for streamflow were rated very good for all three streamgages (table 17). NSE validation statistics for monthly suspended sediment loads were rated good at streamgages 04214500 and 04215000 and unsatisfactory for streamgage 04215500. Most discrepancies between the simulated and the rloadest loads occurred in winter months; especially for the difference between calculated rloadest and simulated loads in February 2019 at the streamgage 04215500. Buffalo Creek had two ice jams at its intersection with NYS Route 277, near streamgage 04214500 on January 25 and February 4, 2019 (USACE, undated a); the later ice jam caused flooding. It is unknown how long exactly each of the ice jams lasted. A field crew documented the second ice jam (fig. 12). The simulated suspended sediment load for February 2019 was over 20,000 metric tons greater than the rloadest load at streamgage 04215500. Validation statistics for simulated total phosphorus loads were rated good for streamgage 04215500 and satisfactory for streamgages 04214500 and 04215000. NSE validation statistics for total nitrogen were rated good for streamgage 04215500 and very good for streamgage 04215000; the NSE validation statistic for total nitrogen for streamgage 04214500 was rated unsatisfactory. Monthly total nitrogen loads for streamgage 04214500 were overestimated in the growing seasons (fig. 10D), as were suspended sediment and total phosphorus loads from July to December 2019.
Photograph of ice jam at U.S. Geological Survey streamgage 04214500 (site 16 in table 1) on Buffalo Creek, New York, on February 5, 2019. Photograph by Elizabeth Nystrom, U.S. Geological Survey.
Canadaway Creek Watershed Model Calibration and Validation
Figure 13 shows the Canadaway Creek watershed model simulated flows and loads compared to the observed streamflow and rloadest suspended sediment and nutrient loads. Flooding on Canadaway Creek on November 5, 2017, caused an overnight bridge closure and evacuations (WIVB-TV, 2017). This corresponds to a peak flow recorded at USGS streamgage 04213376 (USGS, 2016b). Because of this flood, the month of November 2017 had an observed streamflow peak and elevated monthly suspended sediment, total phosphorus, and nitrogen loads higher than previous months (fig. 13). There was a substantial streamflow peak for the month of January 2018. Figure 13 shows that SWAT underestimated the monthly average streamflow, monthly suspended sediment load, and monthly total nutrient loads; this January peak makes up about 73 and 22 percent of the monthly total phosphorus and monthly total nitrogen loads, respectively, for 2018, at streamgage 04213376. The January 2018 peak was removed from the monthly total phosphorus and monthly total nitrogen calibration dataset, that caused an improvement of the model calibration statistics. Table 17 reflects the removal of the January 2018 peak from the calibration dataset. After the removal of the January 2018 calibration data for streamgage 04213376, PBIAS values were rated satisfactory and good for monthly total phosphorus and total nitrogen loads, respectively; the NSE values were both rated satisfactory for monthly total phosphorus and nitrogen loads.
Graphs comparing Soil and Water Assessment Tool simulated monthly streamflow and loads to observed monthly streamflow or rloadest calculated loads from January 2017 to December 2019 at U.S. Geological Survey streamgage 04213376 on Canadaway Creek, in the Canadaway Creek watershed, New York, of A, Average streamflow; B, Suspended sediment load; C, Total phosphorus load; and D, Total nitrogen load.
For validation of the Canadaway Creek model, NSE values for monthly average streamflow, suspended sediment load, and total phosphorus load were rated unsatisfactory (table 17). The PBIAS value for monthly total nitrogen load was rated very good, but the NSE value for monthly total nitrogen load was rated unsatisfactory. Monthly average streamflow was usually underestimated, monthly total nitrogen load values were underestimated and overestimated, and monthly suspended sediment and monthly total phosphorus loads were usually overestimated (fig. 13).
Cattaraugus Creek Watershed Model Calibration and Validation
The streamgage (04213500) used for calibration of the Cattaraugus Creek watershed is 29.5 km upstream from Lake Erie (fig. 2D). The subbasin upstream from the streamgage encompasses about 78 percent of the entire watershed. Withdrawals from the stream had a small effect (<1 percent) on the flow at this point during the study except when a large discharge was released from a wastewater treatment plant (Cattaraugus STP, identifier NY0025861; table 3) in June 2014.
For calibration of the Cattaraugus Creek watershed model, monthly average streamflow and monthly suspended sediment load were rated very good for NSE and good for PBIAS (table 17). The positive PBIAS value (14.82) for the Cattaraugus Creek model calibration period indicates that the model underestimated streamflow (fig. 14). Calibration NSE was rated satisfactory for monthly total phosphorus load and rated good for monthly total nitrogen load. PBIAS values were rated very good for monthly total phosphorus, monthly total nitrogen, and monthly nitrate as nitrogen loads. Monthly total suspended sediment, monthly total nitrogen, monthly nitrate as nitrogen, and monthly total phosphorus loads also have a calibration PBIAS value of <0, indicating that the model overestimated these constituents in the calibration period.
Graphs comparing Soil and Water Assessment Tool simulated monthly streamflow and loads to observed monthly streamflow or rloadest calculated loads from January 2017 to December 2019 at U.S. Geological Survey streamgage 04213500 on Cattaraugus Creek, in the Cattaraugus Creek watershed, New York, of A, Average streamflow; B, Suspended sediment load; C, Total phosphorus load; D, Nitrate as nitrogen load; and E, Total nitrogen load. SWAT-simulated nitrate as nitrogen loads were calibrated to the rloadest calculated nitrate plus nitrite loads (table 16).
Average monthly streamflow was underestimated, and monthly suspended sediment, monthly total nitrogen, and monthly total phosphorus loads were overestimated during the validation period (table 17). The NSE value was rated very good and the PBIAS value was rated good for streamflow in the validation period. Simulated monthly suspended sediment and monthly total phosphorus loads did not have a relation with the estimated rloadest values; the NSE value was <0, and the PBIAS value was rated unsatisfactory for both constituents. The validation NSE value for total nitrogen was rated good, whereas the PBIAS value for total nitrogen was rated very good. Ice jams were reported on Cattaraugus Creek on January 11, 2018, and February 5, 2019 (USACE, undated a); both were downstream of streamgage 04213500.
Chautauqua Creek Watershed Model Calibration and Validation
The SWAT model of the Chautauqua Creek watershed was not rated with satisfactory calibration statistics for monthly average streamflow (table 17). In the calibration period, the Chautauqua Creek SWAT model underestimated monthly average streamflow in winter and spring from November 2017 to April 2018, whereas the model had agreement between simulated and observed monthly average streamflow during the growing season from May to September 2018 (fig. 15). Because model calibration statistics for monthly average streamflow were unsatisfactory, model calibration of suspended sediment and nutrient loads were not attempted. Model result validation statistics were also not calculated.
Graph comparing observed monthly and Soil and Water Assessment Tool simulated monthly streamflow from January 2017 to December 2019 at U.S. Geological Survey streamgage 04213319 on Chautauqua Creek, in the Chautauqua Creek watershed, New York.
Crooked Brook Watershed Model Calibration and Validation
Daily suspended sediment and nutrient loads were not available at Crooked Brook watershed because of the lack of daily streamflow data. As daily streamflow data were not available, monthly loads could not be calculated with rloadest. Because of this, the Crooked Brook watershed model was not calibrated. Instead of traditional model calibration, model calibration parameters were transferred from the adjacent Canadaway Creek watershed model, thus the model of the Crooked Brook watershed uses the same calibration parameters as the Canadaway Creek watershed model shown in table 14. Crooked Brook and other uncalibrated models are not included in table 17. Figure 16 depicts the instantaneous streamflow and daily loads. Observed low streamflow and loads appear to match the simulated datasets, but simulated peaks of streamflow and loads often do not match the observed streamflow and rloadest calculated load peaks. The monitoring data most likely did not capture the peak streamflow, which will skew the comparison between simulated and observed streamflow.
Graphs comparing Soil and Water Assessment Tool simulated daily streamflow and loads to observed daily streamflow or rloadest calculated loads from January 2017 to December 2019 at U.S. Geological Survey streamgage 0421338405 on Crooked Brook, in the Crooked Brook watershed, New York, of A, Instantaneous streamflow; B, Daily suspended sediment load; C, Daily total phosphorus load; D, Daily nitrate as nitrogen load; and E, Daily total nitrogen load. SWAT-simulated nitrate as nitrogen loads were calibrated to the rloadest calculated nitrate plus nitrite loads (table 16).
Eighteenmile Creek Watershed Model Calibration and Validation
The Eighteenmile Creek watershed SWAT model was calibrated at the two streamgages within the watershed (table 17). Following are the calibration statistics for the South Branch Eighteenmile Creek at USGS streamgage 04214231. Calibration NSE values for monthly average streamflow and monthly total nitrogen load were rated good. Monthly suspended sediment loads were not available at streamgage 04214231 (sediment parameters were calibrated at USGS streamgage 0421422210 and these parameters were used for the entire watershed). Calibration NSE values for monthly total phosphorus and nitrate as nitrogen loads at streamgage 04214231 were rated unsatisfactory, but the calibration PBIAS value for monthly total phosphorus load was rated very good. Streamflow and nitrate as nitrogen had satisfactory PBIAS ratings for the calibration period. Calibration NSE and PBIAS for total nitrogen were rated good and very good, respectively, at streamgage 04214231.
In the calibration period, total phosphorus was overestimated at USGS streamgage 04214231 (PBIAS<0; table 17). The SWAT model overestimated total phosphorus for most months in the calibration period (fig. 17B); February 2018 total phosphorus loads were overestimated by about 2,000 kg in the calibration period. Monthly nitrate as nitrogen and total nitrogen loads were underestimated during most months during winter of 2018 to 2019, similar to the underestimated streamflow (figs. 17A, C, D).
Graphs comparing Soil and Water Assessment Tool simulated to observed or rloadest calculated monthly measurements from January 2017 to December 2019 at U.S. Geological Survey streamgage 04214231 on South Branch Eighteenmile Creek, in the Eighteenmile Creek watershed, New York, of A, Average streamflow; B, Total phosphorus load; C, Nitrate as nitrogen load; and D, Total nitrogen load.
The following are calibration statistics for the SWAT model of Eighteenmile Creek at USGS streamgage 0421422210, which drains 159 km2 on the eastern side of the watershed, approximately half of the watershed (table 1; fig. 2G). The NSE and PBIAS values for streamflow were rated very good and satisfactory, respectively (table 17). Monthly suspended sediment and total nitrogen loads had calibration NSE values that were rated satisfactory and calibration PBIAS values that were rated very good, whereas monthly total phosphorus and nitrate as nitrogen loads have calibration NSE values what were rated good and calibration PBIAS values that were rated very good. Simulated monthly average streamflow closely matched observed monthly average streamflow during October 2017 and November 2017 (fig. 18A) in the calibration period, whereas the other SWAT models for the study watersheds tended to underestimate streamflow during November 2017. The following winter, however, simulated monthly average streamflow did not match observed streamflow as closely. Simulated monthly total phosphorus and monthly total nitrogen were overestimated by the model (PBIAS<0). Late spring, in both the calibration and validation periods, had the most discrepancies in the simulated total phosphorus and total nitrogen loads compared to the rloadest values (figs. 17C, E).
Graphs comparing Soil and Water Assessment Tool simulated to observed or rloadest calculated monthly measurements from January 2017 to December 2019 at U.S. Geological Survey streamgage 0421422210 on Eighteenmile Creek, in the Eighteenmile Creek watershed, New York, of A, Average streamflow; B, Suspended sediment load; C, Total phosphorus load; D, Nitrate as nitrogen load; and E, Total nitrogen load. SWAT-simulated nitrate as nitrogen loads were calibrated to the rloadest calculated nitrate plus nitrite loads (table 16).
In the validation period for the SWAT models at both Eighteenmile Creek watershed streamgages, all NSE values were rated unsatisfactory, except for monthly average streamflow at streamgage 0421422210 (table 17). Only monthly average streamflow, monthly suspended sediment load, and monthly nitrate as nitrogen load at streamgage 0421422210 have R2 values that were rated satisfactory. PBIAS values for the validation period at both streamgages were all rated from satisfactory to very good. At streamgage 0421422210, PBIAS values were rated good for monthly average streamflow, very good for monthly suspended sediment and nitrate as nitrogen loads, and satisfactory for monthly total phosphorus and monthly total nitrogen loads. PBIAS values at streamgage 04214231 were rated satisfactory for monthly streamflow, very good for monthly total phosphorus and total nitrogen, and good for nitrate as nitrogen. Most constituents were underestimated in the validation period (PBIAS>0), but monthly total nitrogen was overestimated at streamgage 04214231, and monthly suspended sediment, total phosphorus, and total nitrogen were overestimated at streamgage 0421422210 (PBIAS<0).
Walnut Creek Watershed Model Calibration and Validation
The Walnut Creek watershed model was calibrated to two streamgages (table 17). At both streamgages, calibration NSE values were rated good and calibration PBIAS values were rated very good for streamflow. Because of the lack of statistically significant rloadest models (table 16), monthly suspended sediment and nitrate as nitrogen loads were not calibrated at the Silver Creek streamgage (streamgage 04213394), and monthly total phosphorus loads were not calibrated at the Walnut Creek streamgage (streamgage 04213401). At streamgage 04213394, monthly total phosphorus and total nitrogen loads for both NSE and PBIAS calibration values were rated very good. At streamgage 04213401, the calibration NSE value and the PBIAS value for nitrate as nitrogen was rated unsatisfactory and satisfactory, respectively. Nitrate as nitrogen loads were underestimated for most months at streamgage 04213401 in both the calibration and validation periods (fig. 19C). Ice jams downstream from the streamgages were reported on both Silver and Walnut Creeks on January 11, 2018 (USACE, undated a), which may affect the accuracy of the rloadest load calculations. It is possible that the ice jam caused the large peak in the rloadest monthly suspended sediment load for January 2018 (fig. 19B), where SWAT underestimated the simulated monthly suspended sediment load by over 10,000 metric tons. Similarly, simulated total phosphorus and total nitrogen loads were underestimated at streamgage 04213394 in comparison to th